SlideShare ist ein Scribd-Unternehmen logo
1 von 24
Downloaden Sie, um offline zu lesen
Xen PV Performance Status
    and Optimization Opportunities

           Author: Zhidong Yu
           Contributor: Ronghui Duan, Yuan Zhou, Jian Zhang

           Intel Corporation




1
Disclaimers
    INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE, EXPRESS OR
    IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT
    AS PROVIDED IN INTEL'S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY
    WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL
    PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY,
    OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.
    A "Mission Critical Application" is any application in which failure of the Intel Product could result, directly or indirectly, in
    personal injury or death. SHOULD YOU PURCHASE OR USE INTEL'S PRODUCTS FOR ANY SUCH MISSION CRITICAL
    APPLICATION, YOU SHALL INDEMNIFY AND HOLD INTEL AND ITS SUBSIDIARIES, SUBCONTRACTORS AND AFFILIATES, AND
    THE DIRECTORS, OFFICERS, AND EMPLOYEES OF EACH, HARMLESS AGAINST ALL CLAIMS COSTS, DAMAGES, AND EXPENSES
    AND REASONABLE ATTORNEYS' FEES ARISING OUT OF, DIRECTLY OR INDIRECTLY, ANY CLAIM OF PRODUCT LIABILITY,
    PERSONAL INJURY, OR DEATH ARISING IN ANY WAY OUT OF SUCH MISSION CRITICAL APPLICATION, WHETHER OR NOT
    INTEL OR ITS SUBCONTRACTOR WAS NEGLIGENT IN THE DESIGN, MANUFACTURE, OR WARNING OF THE INTEL PRODUCT OR
    ANY OF ITS PARTS.
    Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the
    absence or characteristics of any features or instructions marked "reserved" or "undefined". Intel reserves these for future
    definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The
    information here is subject to change without notice. Do not finalize a design with this information.
    The products described in this document may contain design defects or errors known as errata which may cause the product to
    deviate from published specifications. Current characterized errata are available on request.
    Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order.
    Copies of documents which have an order number and are referenced in this document, or other Intel literature, may be
    obtained by calling 1-800-548-4725, or go to: http://www.intel.com/design/literature.htm%20
    This document contains information on products in the design phase of development.
    Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.
    *Other names and brands may be claimed as the property of others.
    Copyright © 2012 Intel Corporation. All rights reserved.




2
Agenda

    • HVM vs. PV guest comparison

    • PV disk driver performance and optimizations




3
HVM vs. PV Guest – SPECjbb (1/2)
                               Medium VM                                                                               Large VM




    Source: Intel                                                                      Source: Intel
                        1 VM with 4 vCPUs                                                                    1 VM with 12 vCPUs
               floating on 4 pCPUs (2 cores w/ HT)                                                  floating on 12 pCPUs (6 cores w/ HT)
                        6GB JVM heap size                                                                    18GB JVM heap size



       • SPECjbb is a CPU and memory intensive workload.

       • HVM is 3%~6% better than PV guest.


      Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
      SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
4     factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
      purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
HVM vs. PV Guest – SPECjbb (2/2)

                      Metric       PV Guest HVM Guest HVM/PV
           PathLength                 68,473   66,408   0.97x
            Guest                     66,817   65,542   0.98x
            VMM                        1,656      866   0.52x
           VMExits per second            -     12,650
           hypercalls per second     342,736   65,652   0.19x
            iret                     301,703   15,161   0.05x
            … (the rest)              41,034   50,490   1.23x

    • The major cause is that PV guest has higher Path Length (#
      of instructions per transaction) than HVM
     • in both guest and VMM
     • VMExits are cheaper than hypercalls in this case


             Cheaper VMExits lead to less overhead in Xen
5
HVM vs. PV Guest – SysBench (1/2)




                                         * Xen HVM uses PV driver                                          Source: Intel



     HVM outperforms PV guest by 39%.




    Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
    SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
6   factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
    purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
HVM vs. PV Guest – SysBench (2/2)
                            Metric            PV Guest     HVM      HVM/PV
                   Throughput (txn per sec)       1,730       2,367   1.37x
                   CPU %                           20.9        24.2   1.16x
                   PathLength                 4,852,609   4,821,256   0.99x
                   CPI                             1.61        1.37   0.85x
                   L1 Data Cache MPKI             17.99       14.54   0.81x
                   L2 Cache Data MPKI              4.86        4.23   0.87x
                   L3 Cache Data MPKI              1.02        1.00   0.98x
                   DTLB Load MPKI                  3.48        1.61   0.46x

    There are two reasons for the higher performance of HVM:

    •       More CPU cycles could be utilized

    •       Lower CPI makes the CPU cycles be used more efficiently
        •    Much lower cache misses              PathLength = # of instructions per transaction
                                                  CPI = Cycles per Instruction
                                                  MPKI = Misses per 1000 instructions


7
                               Better Cache Locality in HVM
Agenda

    • HVM vs. PV guest comparison

    • PV disk driver performance and optimizations




8
Big Gap between Bare-metal and PV Disk




                                         Source: Intel



     HVM+PV Disk has 26% lower bandwidth than bare-metal.

     • and much higher CPU utilization.



    Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
    SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
9   factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
    purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
Per-device blkif_io_lock

     Multiple PV disks were used to achieve a high b/w in our test.

     We found the global blkif_io_lock in the front-end seems to
     be the performance bottleneck.

     • Very high CPU utilization in the guest kernel.

     Luckily, this is a known issue and has been fixed already by
     Amazon:

     • Replace the global lock by per-device lock.

     • Commit (March 2012):
       http://git.kernel.org/?p=linux/kernel/git/axboe/linux-
       block.git;a=commit;h=3467811e26660eb46bc655234573d22d6876d5f9




10
Result of Per-device-lock




                   Source: Intel


     However there is no performance improvement.

     • Though it did reduce the overhead (CPU utilization).




11
I/O Merging Issue

     Our further analysis showed the major issue is the I/O request
     cannot get adequate merging in sequential I/O case.

     • A 64KB request in guest became a 44KB request plus a 20KB
       request in dom0.

     The root cause of the problem is a pre-defined MAX limitation in
     the I/O Ring related data structures:

     • #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11

     • Limits the max number of scatter-gather segments in a PV I/O
       request sent from front-end to back-end

     Simply setting the limit to a larger value may lead to new issue:

     • Too few entries accommodated by the ring, which may bring
       other performance issue.

12
                         Details in the next page.
define a
     BLKIF_MAX_SEGMENTS_PER_REQUEST
                                                                                     4KB ring
     File: linux/include/xen/interface/io/blkif.h
     DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response);

     struct blkif_request { refers to
              uint8_t         operation;    /* BLKIF_OP_???                           */
              uint8_t         nr_segments; /* number of segments                      */
              blkif_vdev_t    handle;       /* only for read/write requests           */
              uint64_t        id;           /* private guest value, echoed in resp    */
              union {
                        struct blkif_request_rw rw;
                        struct blkif_request_discard discard;
              } u;
     };             refers to
     struct blkif_request_rw {
              blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
              struct blkif_request_segment {
                        grant_ref_t gref;        /* reference to I/O buffer frame      */
                        /* @first_sect: first sector in frame to transfer (inclusive). */
                        /* @last_sect: last sector in frame to transfer (inclusive).   */
                        uint8_t     first_sect, last_sect;
              } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
                                                                       Determines how
     };                                                                many requests
                                                  limited by          a ring can hold
     #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11

     A segment usually be 4KB I/O block; 11 segments limits the max I/O size to be 44KB
13
Solution: Multi-Page-Ring




     Based on Citrix’s early work.
     •   http://lists.xen.org/archives/html/xen-devel/2012-03/msg00388.html

     Allow to use more than one page for a ring.
     • Still need to increase BLKIF_MAX_SEGMENTS_PER_REQUEST to support more segments in one
       request
     • Request structure size gets increased as a result

     Move seg[BLKIF_MAX_SEGMENTS_PER_REQUEST] out of the request structure and
     put them in a separate ring instead.
     •   More scalable
14
Result of Multi-page-ring




                                              Source: Intel


      Multi-page-ring improved the peak bandwidth by 17% and
      reduced the overhead by 29% at the same time.

      • Still a big gap with the bare-metal.




     Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
     SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
15   factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
     purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
Per-device-lock Again




                                 Source: Intel


      With multi-page-ring, per-device-lock can help get even higher
      b/w and lower overhead.




     Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
     SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
16   factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
     purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
Review the Changes




                  Source: Intel




     Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as
     SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those
17   factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated
     purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
Summary

     HVM outperforms PV guest in both CPU/memory intensive
     workload and disk I/O intensive workload.

     Xen PV disk performance is 26% worse than bare-metal in
     terms of peak bandwidth.

     We proposed a multi-page-ring patch that improves the PV disk
     performance by 24% and reduces the gap with bare-metal to
     single digit.




18
Q&A from the Xen Summit
 Q: There is another queue, the queue of the outstanding requests, did you increase the request queue?
 A: Yes, the data shown in here was with increased outstanding I/O numbers which are large enough to reach
 peak b/w.
 Q: On the PV vs. HVM work, what was the kernel that was being use for SPCjbb, and what distribution?
 A: see backup pages for detailed configurations.
 Q: How many guests where running?
 A: single guest VM.
 Q: Do you have inside into how much the hypervisor contributes to the difference between hypercall and vmexit?
 A: Hypervisor should have no direct contribution to the difference. The difference is majorly from the different
 guest kernels.
 Q: How much of these differences might be helped with APICv?
 A: In this particular case (SPECjbb), APICv won’t help much because SPECjbb is CPU/memory intensive and has
 few external interrupts. APICv may help a lot in an I/O intensive workload.
 What processor was this done on?
 A: see backup pages for detailed configurations.
 Q: How was the data (patch lengh, CPI, cache miss rate) collected?
 A: using Intel in-house tool.
 Q: Was this done with ept?
 A: Yes
 Q: What about shadow?
 A: Did not try it on shadow. Shadow and PV should have similar performance because the use the same page
 table format.
 Q: Use huge pages with specjbb will make a big difference, because it reduces the number of page walks greatly.
 A: guest hugepage was disabled to get a fair comparison between PV and HVM. See backup for configurations.
 Q: Was the test run on rotating media or SSD?
 A:
19    all on SSD.
BACKUP




20
Configurations - SPECjbb
     •       H/W: 2-socket Intel® Xeon® Processor X5660 with 48GB memory (24GB on each node) running at 800MHz.
         •        EIST/Turbo disabled; core c3/c6 disabled.

     •       VMM: Xen 4.1.0
         •        Dom0: Linux 2.6.32 with 2 vCPUs and 4GB memory

     •       Guest O/S: RHEL 6.1
         •        THP is disabled for a fair comparison, because PV guest does not support huge page.
              •       Implies 4KB guest page since we did not reserve huge page explicitly.
         •        With RHEL6.1’s PVOPS kernel, PV disk and PV network are used for HVM
              •       It does not matter much for this particular workload (SPECjbb has almost zero disk/network activities)

     •       VM Configurations:
         •        Case 1: 4 vCPUs + 8GB memory
              •      pinned to 4 pCPUs as a group (floating)
              •      the 4 pCPUs share 2 cores (HT enabled) in socket 0
         •        Case 2: 12 vCPUs + 19GB memory
              •      pinned to 12 pCPUs as a group (floating)
              •      the 12 pCPUs share the socket 0 (HT enabled)

     •       Workload Configurations:
         •        jdk1.6.0_27/bin/java -server –Xmx<N>g –Xms<N>g -XX:+UseParallelOldGC
              •      Case 1: 6GB heap size; Case 2: 18GB heap size

     •       Workload methodology:
         •        Case 1: 3 warehouses for ramp-up and 4 warehouses for steady state.
         •        Case 2: 11 warehouses for ramp-up and 12 warehouses for steady state.

     •       10 minutes ramp-up, and 15 minutes steady state.
         •        Used 60 minutes steady state for data collection.


21
Configurations - SysBench
     H/W: 2-socket Intel® Xeon® Processor E5-2680 with 32GB DDR3 memory running at
     1333MHz. 1x Intel SSD for MySQL data.

     VMM: Xen 4.1.2
     • Dom0: Linux 3.2.9

     Guest O/S: RHEL 6.2
     • With RHEL6.2’s PVOPS kernel, PV disk and PV network were used in both HVM mode and
       PV guest mode.

     VM Configurations:
     • #vCPUs = #pCPUs; 4GB memory
     • The data disk was assigned to VM in ‘PHY:’ mode.

     Workload Configurations:
     • SysBench: prepared 40 million rows ~9.2GB data
     • MySQL: set innodb buffer pool size to 2GB and use O_DIRECT as flush mode.
     • SysBench Read-only test: sysbench --mysql-db=sbtest --max-requests=0 --test=oltp --
       mysql-table-engine=innodb --oltp-table-size=40000000 --mysql-user=root --mysql-
       host=localhost --mysql-socket=/var/lib/mysql/mysql.sock --mysql-password=123456 --
       num-threads=$vCPU_num --max-time=300 --oltp-read-only=on run




22
Configurations – PV Disk
     Hardware:

     • 2-socket Intel® Xeon® Processor X5660 with 48GB memory

     • HBA: LSI 9205

     • Storage: 2 disk bays with 4 Intel SSD disks in each.

     Software:

     • Hypervisor: Xen 4.1.2

     • Dom0: Linux 3.3.6 (8 vCPUs pinned to 8 pCPUs)

     • Guest O/S: Linux 3.3.6 (8 vCPUs pinned to 8 pCPUs; 2GB memory)
       • Virtual disks are configured as ‘PHY:’, i.e., no caching in dom0

     • Workload: aio-stress
       • Accessing disks in O_DIRECT mode, i.e, no caching in guest.
       • Using 64KB sequential I/O pattern




23
Xen PV Performance Status and Optimization Opportunities

Weitere ähnliche Inhalte

Was ist angesagt?

Cloud Computing Hypervisors and Comparison Xen KVM
Cloud Computing Hypervisors and Comparison Xen KVM Cloud Computing Hypervisors and Comparison Xen KVM
Cloud Computing Hypervisors and Comparison Xen KVM cloudresearcher
 
Securing Your Cloud With the Xen Hypervisor by Russell Pavlicek
Securing Your Cloud With the Xen Hypervisor by Russell PavlicekSecuring Your Cloud With the Xen Hypervisor by Russell Pavlicek
Securing Your Cloud With the Xen Hypervisor by Russell Pavlicekbuildacloud
 
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCP
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCPLinuxcon EU : Virtualization in the Cloud featuring Xen and XCP
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCPThe Linux Foundation
 
12 christian ferber xen_server_advanced
12 christian ferber xen_server_advanced12 christian ferber xen_server_advanced
12 christian ferber xen_server_advancedDigicomp Academy AG
 
S4 xen hypervisor_20080622
S4 xen hypervisor_20080622S4 xen hypervisor_20080622
S4 xen hypervisor_20080622Todd Deshane
 
Xen and Client Virtualization: the case of XenClient XT
Xen and Client Virtualization: the case of XenClient XTXen and Client Virtualization: the case of XenClient XT
Xen and Client Virtualization: the case of XenClient XTThe Linux Foundation
 
Rootlinux17: An introduction to Xen Project Virtualisation
Rootlinux17:  An introduction to Xen Project VirtualisationRootlinux17:  An introduction to Xen Project Virtualisation
Rootlinux17: An introduction to Xen Project VirtualisationThe Linux Foundation
 
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,Pavlicek
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,PavlicekXen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,Pavlicek
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,PavlicekThe Linux Foundation
 
Virtualization with KVM (Kernel-based Virtual Machine)
Virtualization with KVM (Kernel-based Virtual Machine)Virtualization with KVM (Kernel-based Virtual Machine)
Virtualization with KVM (Kernel-based Virtual Machine)Novell
 
OSSEU18: NVDIMM and Virtualization - George Dunlap, Citrix
OSSEU18: NVDIMM and Virtualization  - George Dunlap, CitrixOSSEU18: NVDIMM and Virtualization  - George Dunlap, Citrix
OSSEU18: NVDIMM and Virtualization - George Dunlap, CitrixThe Linux Foundation
 
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...The Linux Foundation
 
LFNW2014 Advanced Security Features of Xen Project Hypervisor
LFNW2014 Advanced Security Features of Xen Project HypervisorLFNW2014 Advanced Security Features of Xen Project Hypervisor
LFNW2014 Advanced Security Features of Xen Project HypervisorThe Linux Foundation
 
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)The Linux Foundation
 

Was ist angesagt? (20)

Cloud Computing Hypervisors and Comparison Xen KVM
Cloud Computing Hypervisors and Comparison Xen KVM Cloud Computing Hypervisors and Comparison Xen KVM
Cloud Computing Hypervisors and Comparison Xen KVM
 
Securing Your Cloud With the Xen Hypervisor by Russell Pavlicek
Securing Your Cloud With the Xen Hypervisor by Russell PavlicekSecuring Your Cloud With the Xen Hypervisor by Russell Pavlicek
Securing Your Cloud With the Xen Hypervisor by Russell Pavlicek
 
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCP
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCPLinuxcon EU : Virtualization in the Cloud featuring Xen and XCP
Linuxcon EU : Virtualization in the Cloud featuring Xen and XCP
 
12 christian ferber xen_server_advanced
12 christian ferber xen_server_advanced12 christian ferber xen_server_advanced
12 christian ferber xen_server_advanced
 
S4 xen hypervisor_20080622
S4 xen hypervisor_20080622S4 xen hypervisor_20080622
S4 xen hypervisor_20080622
 
Xen and Client Virtualization: the case of XenClient XT
Xen and Client Virtualization: the case of XenClient XTXen and Client Virtualization: the case of XenClient XT
Xen and Client Virtualization: the case of XenClient XT
 
Aplura virtualization slides
Aplura virtualization slidesAplura virtualization slides
Aplura virtualization slides
 
µ-Xen
µ-Xenµ-Xen
µ-Xen
 
Rootlinux17: An introduction to Xen Project Virtualisation
Rootlinux17:  An introduction to Xen Project VirtualisationRootlinux17:  An introduction to Xen Project Virtualisation
Rootlinux17: An introduction to Xen Project Virtualisation
 
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,Pavlicek
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,PavlicekXen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,Pavlicek
Xen, XenServer, and XAPI: What’s the Difference?-XPUS13 Bulpin,Pavlicek
 
XS Boston 2008 Memory Overcommit
XS Boston 2008 Memory OvercommitXS Boston 2008 Memory Overcommit
XS Boston 2008 Memory Overcommit
 
Virtualization with KVM (Kernel-based Virtual Machine)
Virtualization with KVM (Kernel-based Virtual Machine)Virtualization with KVM (Kernel-based Virtual Machine)
Virtualization with KVM (Kernel-based Virtual Machine)
 
OSSEU18: NVDIMM and Virtualization - George Dunlap, Citrix
OSSEU18: NVDIMM and Virtualization  - George Dunlap, CitrixOSSEU18: NVDIMM and Virtualization  - George Dunlap, Citrix
OSSEU18: NVDIMM and Virtualization - George Dunlap, Citrix
 
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...
XPDS13: In-Guest Mechanism to Strengthen Guest Separation - Philip Tricca, Ci...
 
Why xen slides
Why xen slidesWhy xen slides
Why xen slides
 
LFNW2014 Advanced Security Features of Xen Project Hypervisor
LFNW2014 Advanced Security Features of Xen Project HypervisorLFNW2014 Advanced Security Features of Xen Project Hypervisor
LFNW2014 Advanced Security Features of Xen Project Hypervisor
 
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)
Xen cloud platform v1.1 (given at Build a Cloud Day in Antwerp)
 
XS Boston 2008 Quantitative
XS Boston 2008 QuantitativeXS Boston 2008 Quantitative
XS Boston 2008 Quantitative
 
Xen time machine
Xen time machineXen time machine
Xen time machine
 
XS Boston 2008 XenLoop
XS Boston 2008 XenLoopXS Boston 2008 XenLoop
XS Boston 2008 XenLoop
 

Andere mochten auch

Mirage: extreme specialisation of virtual appliances
Mirage: extreme specialisation of virtual appliancesMirage: extreme specialisation of virtual appliances
Mirage: extreme specialisation of virtual appliancesThe Linux Foundation
 
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCP
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCPOscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCP
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCPThe Linux Foundation
 
Windsor: Domain 0 Disaggregation for XenServer and XCP
	Windsor: Domain 0 Disaggregation for XenServer and XCP	Windsor: Domain 0 Disaggregation for XenServer and XCP
Windsor: Domain 0 Disaggregation for XenServer and XCPThe Linux Foundation
 
Scale11x : Virtualization with Xen and XCP
Scale11x : Virtualization with Xen and XCP Scale11x : Virtualization with Xen and XCP
Scale11x : Virtualization with Xen and XCP The Linux Foundation
 
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServer
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServerUnder the Hood: Open vSwitch & OpenFlow in XCP & XenServer
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServerThe Linux Foundation
 

Andere mochten auch (12)

Openstack Xen and XCP
Openstack Xen and XCPOpenstack Xen and XCP
Openstack Xen and XCP
 
Art of Using Xen at Scale
Art of Using Xen at ScaleArt of Using Xen at Scale
Art of Using Xen at Scale
 
Mirage: extreme specialisation of virtual appliances
Mirage: extreme specialisation of virtual appliancesMirage: extreme specialisation of virtual appliances
Mirage: extreme specialisation of virtual appliances
 
Engaging the xen community
Engaging the xen communityEngaging the xen community
Engaging the xen community
 
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCP
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCPOscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCP
Oscon 2012 : From Datacenter to the Cloud - Featuring Xen and XCP
 
Windsor: Domain 0 Disaggregation for XenServer and XCP
	Windsor: Domain 0 Disaggregation for XenServer and XCP	Windsor: Domain 0 Disaggregation for XenServer and XCP
Windsor: Domain 0 Disaggregation for XenServer and XCP
 
Xen and Apache cloudstack
Xen and Apache cloudstack  Xen and Apache cloudstack
Xen and Apache cloudstack
 
Scale11x : Virtualization with Xen and XCP
Scale11x : Virtualization with Xen and XCP Scale11x : Virtualization with Xen and XCP
Scale11x : Virtualization with Xen and XCP
 
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServer
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServerUnder the Hood: Open vSwitch & OpenFlow in XCP & XenServer
Under the Hood: Open vSwitch & OpenFlow in XCP & XenServer
 
Xen @ Google, 2011
Xen @ Google, 2011Xen @ Google, 2011
Xen @ Google, 2011
 
Xen 4.3 Roadmap
Xen 4.3 RoadmapXen 4.3 Roadmap
Xen 4.3 Roadmap
 
µ-Xen
µ-Xenµ-Xen
µ-Xen
 

Ähnlich wie Xen PV Performance Status and Optimization Opportunities

Accelerating Virtual Machine Access with the Storage Performance Development ...
Accelerating Virtual Machine Access with the Storage Performance Development ...Accelerating Virtual Machine Access with the Storage Performance Development ...
Accelerating Virtual Machine Access with the Storage Performance Development ...Michelle Holley
 
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...The Linux Foundation
 
Lakefield: Hybrid Cores in 3D Package
Lakefield: Hybrid Cores in 3D PackageLakefield: Hybrid Cores in 3D Package
Lakefield: Hybrid Cores in 3D Packageinside-BigData.com
 
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTEL
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTELSTATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTEL
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTELThe Linux Foundation
 
Deep Learning Training at Scale: Spring Crest Deep Learning Accelerator
Deep Learning Training at Scale: Spring Crest Deep Learning AcceleratorDeep Learning Training at Scale: Spring Crest Deep Learning Accelerator
Deep Learning Training at Scale: Spring Crest Deep Learning Acceleratorinside-BigData.com
 
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...The Linux Foundation
 
Performance out of the box developers
Performance   out of the box developersPerformance   out of the box developers
Performance out of the box developersMichelle Holley
 
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel Architecture
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel ArchitectureDPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel Architecture
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel ArchitectureJim St. Leger
 
Introduction to container networking in K8s - SDN/NFV London meetup
Introduction to container networking in K8s - SDN/NFV  London meetupIntroduction to container networking in K8s - SDN/NFV  London meetup
Introduction to container networking in K8s - SDN/NFV London meetupHaidee McMahon
 
Spring Hill (NNP-I 1000): Intel's Data Center Inference Chip
Spring Hill (NNP-I 1000): Intel's Data Center Inference ChipSpring Hill (NNP-I 1000): Intel's Data Center Inference Chip
Spring Hill (NNP-I 1000): Intel's Data Center Inference Chipinside-BigData.com
 
Omni path-fabric-software-architecture-overview
Omni path-fabric-software-architecture-overviewOmni path-fabric-software-architecture-overview
Omni path-fabric-software-architecture-overviewDESMOND YUEN
 
Overview of Intel® Omni-Path Architecture
Overview of Intel® Omni-Path ArchitectureOverview of Intel® Omni-Path Architecture
Overview of Intel® Omni-Path ArchitectureIntel® Software
 
How to Boost 100x Performance for Real World Application with Apache Spark-(G...
How to Boost 100x Performance for Real World Application with Apache Spark-(G...How to Boost 100x Performance for Real World Application with Apache Spark-(G...
How to Boost 100x Performance for Real World Application with Apache Spark-(G...Spark Summit
 
LF_DPDK_Accelerate storage service via SPDK
LF_DPDK_Accelerate storage service via SPDK LF_DPDK_Accelerate storage service via SPDK
LF_DPDK_Accelerate storage service via SPDK LF_DPDK
 
2 new hw_features_cat_cod_etc
2 new hw_features_cat_cod_etc2 new hw_features_cat_cod_etc
2 new hw_features_cat_cod_etcvideos
 
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDS
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDSAccelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDS
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDSCeph Community
 
Cheapest Analysis Essay
Cheapest Analysis EssayCheapest Analysis Essay
Cheapest Analysis EssayGloria Moore
 
Quieting noisy neighbor with Intel® Resource Director Technology
Quieting noisy neighbor with Intel® Resource Director TechnologyQuieting noisy neighbor with Intel® Resource Director Technology
Quieting noisy neighbor with Intel® Resource Director TechnologyMichelle Holley
 
hbaseconasia2019 HBase Bucket Cache on Persistent Memory
hbaseconasia2019 HBase Bucket Cache on Persistent Memoryhbaseconasia2019 HBase Bucket Cache on Persistent Memory
hbaseconasia2019 HBase Bucket Cache on Persistent MemoryMichael Stack
 

Ähnlich wie Xen PV Performance Status and Optimization Opportunities (20)

Accelerating Virtual Machine Access with the Storage Performance Development ...
Accelerating Virtual Machine Access with the Storage Performance Development ...Accelerating Virtual Machine Access with the Storage Performance Development ...
Accelerating Virtual Machine Access with the Storage Performance Development ...
 
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...
XPDS16: High-Performance Virtualization for HPC Cloud on Xen - Jun Nakajima &...
 
Lakefield: Hybrid Cores in 3D Package
Lakefield: Hybrid Cores in 3D PackageLakefield: Hybrid Cores in 3D Package
Lakefield: Hybrid Cores in 3D Package
 
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTEL
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTELSTATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTEL
STATUS UPDATE OF COLO PROJECT XIAOWEI YANG, HUAWEI AND WILL AULD, INTEL
 
Deep Learning Training at Scale: Spring Crest Deep Learning Accelerator
Deep Learning Training at Scale: Spring Crest Deep Learning AcceleratorDeep Learning Training at Scale: Spring Crest Deep Learning Accelerator
Deep Learning Training at Scale: Spring Crest Deep Learning Accelerator
 
Mcdram Tutorial
Mcdram TutorialMcdram Tutorial
Mcdram Tutorial
 
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...
XPDS13: Performance Optimization on Xen-based Android Device - Jack Ren, Inte...
 
Performance out of the box developers
Performance   out of the box developersPerformance   out of the box developers
Performance out of the box developers
 
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel Architecture
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel ArchitectureDPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel Architecture
DPDK Summit - 08 Sept 2014 - Intel - Networking Workloads on Intel Architecture
 
Introduction to container networking in K8s - SDN/NFV London meetup
Introduction to container networking in K8s - SDN/NFV  London meetupIntroduction to container networking in K8s - SDN/NFV  London meetup
Introduction to container networking in K8s - SDN/NFV London meetup
 
Spring Hill (NNP-I 1000): Intel's Data Center Inference Chip
Spring Hill (NNP-I 1000): Intel's Data Center Inference ChipSpring Hill (NNP-I 1000): Intel's Data Center Inference Chip
Spring Hill (NNP-I 1000): Intel's Data Center Inference Chip
 
Omni path-fabric-software-architecture-overview
Omni path-fabric-software-architecture-overviewOmni path-fabric-software-architecture-overview
Omni path-fabric-software-architecture-overview
 
Overview of Intel® Omni-Path Architecture
Overview of Intel® Omni-Path ArchitectureOverview of Intel® Omni-Path Architecture
Overview of Intel® Omni-Path Architecture
 
How to Boost 100x Performance for Real World Application with Apache Spark-(G...
How to Boost 100x Performance for Real World Application with Apache Spark-(G...How to Boost 100x Performance for Real World Application with Apache Spark-(G...
How to Boost 100x Performance for Real World Application with Apache Spark-(G...
 
LF_DPDK_Accelerate storage service via SPDK
LF_DPDK_Accelerate storage service via SPDK LF_DPDK_Accelerate storage service via SPDK
LF_DPDK_Accelerate storage service via SPDK
 
2 new hw_features_cat_cod_etc
2 new hw_features_cat_cod_etc2 new hw_features_cat_cod_etc
2 new hw_features_cat_cod_etc
 
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDS
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDSAccelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDS
Accelerating Cassandra Workloads on Ceph with All-Flash PCIE SSDS
 
Cheapest Analysis Essay
Cheapest Analysis EssayCheapest Analysis Essay
Cheapest Analysis Essay
 
Quieting noisy neighbor with Intel® Resource Director Technology
Quieting noisy neighbor with Intel® Resource Director TechnologyQuieting noisy neighbor with Intel® Resource Director Technology
Quieting noisy neighbor with Intel® Resource Director Technology
 
hbaseconasia2019 HBase Bucket Cache on Persistent Memory
hbaseconasia2019 HBase Bucket Cache on Persistent Memoryhbaseconasia2019 HBase Bucket Cache on Persistent Memory
hbaseconasia2019 HBase Bucket Cache on Persistent Memory
 

Mehr von The Linux Foundation

ELC2019: Static Partitioning Made Simple
ELC2019: Static Partitioning Made SimpleELC2019: Static Partitioning Made Simple
ELC2019: Static Partitioning Made SimpleThe Linux Foundation
 
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...The Linux Foundation
 
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...The Linux Foundation
 
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...The Linux Foundation
 
XPDDS19 Keynote: Unikraft Weather Report
XPDDS19 Keynote:  Unikraft Weather ReportXPDDS19 Keynote:  Unikraft Weather Report
XPDDS19 Keynote: Unikraft Weather ReportThe Linux Foundation
 
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...The Linux Foundation
 
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, Xilinx
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, XilinxXPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, Xilinx
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, XilinxThe Linux Foundation
 
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...The Linux Foundation
 
XPDDS19: Memories of a VM Funk - Mihai Donțu, Bitdefender
XPDDS19: Memories of a VM Funk - Mihai Donțu, BitdefenderXPDDS19: Memories of a VM Funk - Mihai Donțu, Bitdefender
XPDDS19: Memories of a VM Funk - Mihai Donțu, BitdefenderThe Linux Foundation
 
OSSJP/ALS19: The Road to Safety Certification: Overcoming Community Challeng...
OSSJP/ALS19:  The Road to Safety Certification: Overcoming Community Challeng...OSSJP/ALS19:  The Road to Safety Certification: Overcoming Community Challeng...
OSSJP/ALS19: The Road to Safety Certification: Overcoming Community Challeng...The Linux Foundation
 
OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...
 OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making... OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...
OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...The Linux Foundation
 
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, Citrix
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, CitrixXPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, Citrix
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, CitrixThe Linux Foundation
 
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltd
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltdXPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltd
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltdThe Linux Foundation
 
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...The Linux Foundation
 
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&D
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&DXPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&D
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&DThe Linux Foundation
 
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM Systems
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM SystemsXPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM Systems
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM SystemsThe Linux Foundation
 
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...The Linux Foundation
 
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...The Linux Foundation
 
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...The Linux Foundation
 
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSE
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSEXPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSE
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSEThe Linux Foundation
 

Mehr von The Linux Foundation (20)

ELC2019: Static Partitioning Made Simple
ELC2019: Static Partitioning Made SimpleELC2019: Static Partitioning Made Simple
ELC2019: Static Partitioning Made Simple
 
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...
XPDDS19: How TrenchBoot is Enabling Measured Launch for Open-Source Platform ...
 
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...
XPDDS19 Keynote: Xen in Automotive - Artem Mygaiev, Director, Technology Solu...
 
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...
XPDDS19 Keynote: Xen Project Weather Report 2019 - Lars Kurth, Director of Op...
 
XPDDS19 Keynote: Unikraft Weather Report
XPDDS19 Keynote:  Unikraft Weather ReportXPDDS19 Keynote:  Unikraft Weather Report
XPDDS19 Keynote: Unikraft Weather Report
 
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...
XPDDS19 Keynote: Secret-free Hypervisor: Now and Future - Wei Liu, Software E...
 
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, Xilinx
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, XilinxXPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, Xilinx
XPDDS19 Keynote: Xen Dom0-less - Stefano Stabellini, Principal Engineer, Xilinx
 
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...
XPDDS19 Keynote: Patch Review for Non-maintainers - George Dunlap, Citrix Sys...
 
XPDDS19: Memories of a VM Funk - Mihai Donțu, Bitdefender
XPDDS19: Memories of a VM Funk - Mihai Donțu, BitdefenderXPDDS19: Memories of a VM Funk - Mihai Donțu, Bitdefender
XPDDS19: Memories of a VM Funk - Mihai Donțu, Bitdefender
 
OSSJP/ALS19: The Road to Safety Certification: Overcoming Community Challeng...
OSSJP/ALS19:  The Road to Safety Certification: Overcoming Community Challeng...OSSJP/ALS19:  The Road to Safety Certification: Overcoming Community Challeng...
OSSJP/ALS19: The Road to Safety Certification: Overcoming Community Challeng...
 
OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...
 OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making... OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...
OSSJP/ALS19: The Road to Safety Certification: How the Xen Project is Making...
 
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, Citrix
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, CitrixXPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, Citrix
XPDDS19: Speculative Sidechannels and Mitigations - Andrew Cooper, Citrix
 
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltd
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltdXPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltd
XPDDS19: Keeping Coherency on Arm: Reborn - Julien Grall, Arm ltd
 
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...
XPDDS19: QEMU PV Backend 'qdevification'... What Does it Mean? - Paul Durrant...
 
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&D
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&DXPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&D
XPDDS19: Status of PCI Emulation in Xen - Roger Pau Monné, Citrix Systems R&D
 
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM Systems
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM SystemsXPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM Systems
XPDDS19: [ARM] OP-TEE Mediator in Xen - Volodymyr Babchuk, EPAM Systems
 
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...
XPDDS19: Bringing Xen to the Masses: The Story of Building a Community-driven...
 
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...
XPDDS19: Will Robots Automate Your Job Away? Streamlining Xen Project Contrib...
 
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...
XPDDS19: Client Virtualization Toolstack in Go - Nick Rosbrook & Brendan Kerr...
 
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSE
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSEXPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSE
XPDDS19: Core Scheduling in Xen - Jürgen Groß, SUSE
 

Kürzlich hochgeladen

Crea il tuo assistente AI con lo Stregatto (open source python framework)
Crea il tuo assistente AI con lo Stregatto (open source python framework)Crea il tuo assistente AI con lo Stregatto (open source python framework)
Crea il tuo assistente AI con lo Stregatto (open source python framework)Commit University
 
AI Fame Rush Review – Virtual Influencer Creation In Just Minutes
AI Fame Rush Review – Virtual Influencer Creation In Just MinutesAI Fame Rush Review – Virtual Influencer Creation In Just Minutes
AI Fame Rush Review – Virtual Influencer Creation In Just MinutesMd Hossain Ali
 
Videogame localization & technology_ how to enhance the power of translation.pdf
Videogame localization & technology_ how to enhance the power of translation.pdfVideogame localization & technology_ how to enhance the power of translation.pdf
Videogame localization & technology_ how to enhance the power of translation.pdfinfogdgmi
 
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...DianaGray10
 
Building AI-Driven Apps Using Semantic Kernel.pptx
Building AI-Driven Apps Using Semantic Kernel.pptxBuilding AI-Driven Apps Using Semantic Kernel.pptx
Building AI-Driven Apps Using Semantic Kernel.pptxUdaiappa Ramachandran
 
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve Decarbonization
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve DecarbonizationUsing IESVE for Loads, Sizing and Heat Pump Modeling to Achieve Decarbonization
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve DecarbonizationIES VE
 
Computer 10: Lesson 10 - Online Crimes and Hazards
Computer 10: Lesson 10 - Online Crimes and HazardsComputer 10: Lesson 10 - Online Crimes and Hazards
Computer 10: Lesson 10 - Online Crimes and HazardsSeth Reyes
 
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCost
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCostKubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCost
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCostMatt Ray
 
Empowering Africa's Next Generation: The AI Leadership Blueprint
Empowering Africa's Next Generation: The AI Leadership BlueprintEmpowering Africa's Next Generation: The AI Leadership Blueprint
Empowering Africa's Next Generation: The AI Leadership BlueprintMahmoud Rabie
 
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDE
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDEADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDE
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDELiveplex
 
UiPath Community: AI for UiPath Automation Developers
UiPath Community: AI for UiPath Automation DevelopersUiPath Community: AI for UiPath Automation Developers
UiPath Community: AI for UiPath Automation DevelopersUiPathCommunity
 
Building Your Own AI Instance (TBLC AI )
Building Your Own AI Instance (TBLC AI )Building Your Own AI Instance (TBLC AI )
Building Your Own AI Instance (TBLC AI )Brian Pichman
 
Bird eye's view on Camunda open source ecosystem
Bird eye's view on Camunda open source ecosystemBird eye's view on Camunda open source ecosystem
Bird eye's view on Camunda open source ecosystemAsko Soukka
 
Introduction to Matsuo Laboratory (ENG).pptx
Introduction to Matsuo Laboratory (ENG).pptxIntroduction to Matsuo Laboratory (ENG).pptx
Introduction to Matsuo Laboratory (ENG).pptxMatsuo Lab
 
Cybersecurity Workshop #1.pptx
Cybersecurity Workshop #1.pptxCybersecurity Workshop #1.pptx
Cybersecurity Workshop #1.pptxGDSC PJATK
 
Meet the new FSP 3000 M-Flex800™
Meet the new FSP 3000 M-Flex800™Meet the new FSP 3000 M-Flex800™
Meet the new FSP 3000 M-Flex800™Adtran
 
VoIP Service and Marketing using Odoo and Asterisk PBX
VoIP Service and Marketing using Odoo and Asterisk PBXVoIP Service and Marketing using Odoo and Asterisk PBX
VoIP Service and Marketing using Odoo and Asterisk PBXTarek Kalaji
 
UiPath Platform: The Backend Engine Powering Your Automation - Session 1
UiPath Platform: The Backend Engine Powering Your Automation - Session 1UiPath Platform: The Backend Engine Powering Your Automation - Session 1
UiPath Platform: The Backend Engine Powering Your Automation - Session 1DianaGray10
 
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...Will Schroeder
 

Kürzlich hochgeladen (20)

Crea il tuo assistente AI con lo Stregatto (open source python framework)
Crea il tuo assistente AI con lo Stregatto (open source python framework)Crea il tuo assistente AI con lo Stregatto (open source python framework)
Crea il tuo assistente AI con lo Stregatto (open source python framework)
 
AI Fame Rush Review – Virtual Influencer Creation In Just Minutes
AI Fame Rush Review – Virtual Influencer Creation In Just MinutesAI Fame Rush Review – Virtual Influencer Creation In Just Minutes
AI Fame Rush Review – Virtual Influencer Creation In Just Minutes
 
Videogame localization & technology_ how to enhance the power of translation.pdf
Videogame localization & technology_ how to enhance the power of translation.pdfVideogame localization & technology_ how to enhance the power of translation.pdf
Videogame localization & technology_ how to enhance the power of translation.pdf
 
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...
Connector Corner: Extending LLM automation use cases with UiPath GenAI connec...
 
Building AI-Driven Apps Using Semantic Kernel.pptx
Building AI-Driven Apps Using Semantic Kernel.pptxBuilding AI-Driven Apps Using Semantic Kernel.pptx
Building AI-Driven Apps Using Semantic Kernel.pptx
 
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve Decarbonization
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve DecarbonizationUsing IESVE for Loads, Sizing and Heat Pump Modeling to Achieve Decarbonization
Using IESVE for Loads, Sizing and Heat Pump Modeling to Achieve Decarbonization
 
Computer 10: Lesson 10 - Online Crimes and Hazards
Computer 10: Lesson 10 - Online Crimes and HazardsComputer 10: Lesson 10 - Online Crimes and Hazards
Computer 10: Lesson 10 - Online Crimes and Hazards
 
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCost
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCostKubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCost
KubeConEU24-Monitoring Kubernetes and Cloud Spend with OpenCost
 
Empowering Africa's Next Generation: The AI Leadership Blueprint
Empowering Africa's Next Generation: The AI Leadership BlueprintEmpowering Africa's Next Generation: The AI Leadership Blueprint
Empowering Africa's Next Generation: The AI Leadership Blueprint
 
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDE
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDEADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDE
ADOPTING WEB 3 FOR YOUR BUSINESS: A STEP-BY-STEP GUIDE
 
UiPath Community: AI for UiPath Automation Developers
UiPath Community: AI for UiPath Automation DevelopersUiPath Community: AI for UiPath Automation Developers
UiPath Community: AI for UiPath Automation Developers
 
Building Your Own AI Instance (TBLC AI )
Building Your Own AI Instance (TBLC AI )Building Your Own AI Instance (TBLC AI )
Building Your Own AI Instance (TBLC AI )
 
Bird eye's view on Camunda open source ecosystem
Bird eye's view on Camunda open source ecosystemBird eye's view on Camunda open source ecosystem
Bird eye's view on Camunda open source ecosystem
 
201610817 - edge part1
201610817 - edge part1201610817 - edge part1
201610817 - edge part1
 
Introduction to Matsuo Laboratory (ENG).pptx
Introduction to Matsuo Laboratory (ENG).pptxIntroduction to Matsuo Laboratory (ENG).pptx
Introduction to Matsuo Laboratory (ENG).pptx
 
Cybersecurity Workshop #1.pptx
Cybersecurity Workshop #1.pptxCybersecurity Workshop #1.pptx
Cybersecurity Workshop #1.pptx
 
Meet the new FSP 3000 M-Flex800™
Meet the new FSP 3000 M-Flex800™Meet the new FSP 3000 M-Flex800™
Meet the new FSP 3000 M-Flex800™
 
VoIP Service and Marketing using Odoo and Asterisk PBX
VoIP Service and Marketing using Odoo and Asterisk PBXVoIP Service and Marketing using Odoo and Asterisk PBX
VoIP Service and Marketing using Odoo and Asterisk PBX
 
UiPath Platform: The Backend Engine Powering Your Automation - Session 1
UiPath Platform: The Backend Engine Powering Your Automation - Session 1UiPath Platform: The Backend Engine Powering Your Automation - Session 1
UiPath Platform: The Backend Engine Powering Your Automation - Session 1
 
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...
Apres-Cyber - The Data Dilemma: Bridging Offensive Operations and Machine Lea...
 

Xen PV Performance Status and Optimization Opportunities

  • 1. Xen PV Performance Status and Optimization Opportunities Author: Zhidong Yu Contributor: Ronghui Duan, Yuan Zhou, Jian Zhang Intel Corporation 1
  • 2. Disclaimers INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL'S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. A "Mission Critical Application" is any application in which failure of the Intel Product could result, directly or indirectly, in personal injury or death. SHOULD YOU PURCHASE OR USE INTEL'S PRODUCTS FOR ANY SUCH MISSION CRITICAL APPLICATION, YOU SHALL INDEMNIFY AND HOLD INTEL AND ITS SUBSIDIARIES, SUBCONTRACTORS AND AFFILIATES, AND THE DIRECTORS, OFFICERS, AND EMPLOYEES OF EACH, HARMLESS AGAINST ALL CLAIMS COSTS, DAMAGES, AND EXPENSES AND REASONABLE ATTORNEYS' FEES ARISING OUT OF, DIRECTLY OR INDIRECTLY, ANY CLAIM OF PRODUCT LIABILITY, PERSONAL INJURY, OR DEATH ARISING IN ANY WAY OUT OF SUCH MISSION CRITICAL APPLICATION, WHETHER OR NOT INTEL OR ITS SUBCONTRACTOR WAS NEGLIGENT IN THE DESIGN, MANUFACTURE, OR WARNING OF THE INTEL PRODUCT OR ANY OF ITS PARTS. Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the absence or characteristics of any features or instructions marked "reserved" or "undefined". Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not finalize a design with this information. The products described in this document may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order. Copies of documents which have an order number and are referenced in this document, or other Intel literature, may be obtained by calling 1-800-548-4725, or go to: http://www.intel.com/design/literature.htm%20 This document contains information on products in the design phase of development. Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries. *Other names and brands may be claimed as the property of others. Copyright © 2012 Intel Corporation. All rights reserved. 2
  • 3. Agenda • HVM vs. PV guest comparison • PV disk driver performance and optimizations 3
  • 4. HVM vs. PV Guest – SPECjbb (1/2) Medium VM Large VM Source: Intel Source: Intel 1 VM with 4 vCPUs 1 VM with 12 vCPUs floating on 4 pCPUs (2 cores w/ HT) floating on 12 pCPUs (6 cores w/ HT) 6GB JVM heap size 18GB JVM heap size • SPECjbb is a CPU and memory intensive workload. • HVM is 3%~6% better than PV guest. Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 4 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 5. HVM vs. PV Guest – SPECjbb (2/2) Metric PV Guest HVM Guest HVM/PV PathLength 68,473 66,408 0.97x Guest 66,817 65,542 0.98x VMM 1,656 866 0.52x VMExits per second - 12,650 hypercalls per second 342,736 65,652 0.19x iret 301,703 15,161 0.05x … (the rest) 41,034 50,490 1.23x • The major cause is that PV guest has higher Path Length (# of instructions per transaction) than HVM • in both guest and VMM • VMExits are cheaper than hypercalls in this case Cheaper VMExits lead to less overhead in Xen 5
  • 6. HVM vs. PV Guest – SysBench (1/2) * Xen HVM uses PV driver Source: Intel HVM outperforms PV guest by 39%. Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 6 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 7. HVM vs. PV Guest – SysBench (2/2) Metric PV Guest HVM HVM/PV Throughput (txn per sec) 1,730 2,367 1.37x CPU % 20.9 24.2 1.16x PathLength 4,852,609 4,821,256 0.99x CPI 1.61 1.37 0.85x L1 Data Cache MPKI 17.99 14.54 0.81x L2 Cache Data MPKI 4.86 4.23 0.87x L3 Cache Data MPKI 1.02 1.00 0.98x DTLB Load MPKI 3.48 1.61 0.46x There are two reasons for the higher performance of HVM: • More CPU cycles could be utilized • Lower CPI makes the CPU cycles be used more efficiently • Much lower cache misses PathLength = # of instructions per transaction CPI = Cycles per Instruction MPKI = Misses per 1000 instructions 7 Better Cache Locality in HVM
  • 8. Agenda • HVM vs. PV guest comparison • PV disk driver performance and optimizations 8
  • 9. Big Gap between Bare-metal and PV Disk Source: Intel HVM+PV Disk has 26% lower bandwidth than bare-metal. • and much higher CPU utilization. Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 9 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 10. Per-device blkif_io_lock Multiple PV disks were used to achieve a high b/w in our test. We found the global blkif_io_lock in the front-end seems to be the performance bottleneck. • Very high CPU utilization in the guest kernel. Luckily, this is a known issue and has been fixed already by Amazon: • Replace the global lock by per-device lock. • Commit (March 2012): http://git.kernel.org/?p=linux/kernel/git/axboe/linux- block.git;a=commit;h=3467811e26660eb46bc655234573d22d6876d5f9 10
  • 11. Result of Per-device-lock Source: Intel However there is no performance improvement. • Though it did reduce the overhead (CPU utilization). 11
  • 12. I/O Merging Issue Our further analysis showed the major issue is the I/O request cannot get adequate merging in sequential I/O case. • A 64KB request in guest became a 44KB request plus a 20KB request in dom0. The root cause of the problem is a pre-defined MAX limitation in the I/O Ring related data structures: • #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11 • Limits the max number of scatter-gather segments in a PV I/O request sent from front-end to back-end Simply setting the limit to a larger value may lead to new issue: • Too few entries accommodated by the ring, which may bring other performance issue. 12 Details in the next page.
  • 13. define a BLKIF_MAX_SEGMENTS_PER_REQUEST 4KB ring File: linux/include/xen/interface/io/blkif.h DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response); struct blkif_request { refers to uint8_t operation; /* BLKIF_OP_??? */ uint8_t nr_segments; /* number of segments */ blkif_vdev_t handle; /* only for read/write requests */ uint64_t id; /* private guest value, echoed in resp */ union { struct blkif_request_rw rw; struct blkif_request_discard discard; } u; }; refers to struct blkif_request_rw { blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ struct blkif_request_segment { grant_ref_t gref; /* reference to I/O buffer frame */ /* @first_sect: first sector in frame to transfer (inclusive). */ /* @last_sect: last sector in frame to transfer (inclusive). */ uint8_t first_sect, last_sect; } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; Determines how }; many requests limited by a ring can hold #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11 A segment usually be 4KB I/O block; 11 segments limits the max I/O size to be 44KB 13
  • 14. Solution: Multi-Page-Ring Based on Citrix’s early work. • http://lists.xen.org/archives/html/xen-devel/2012-03/msg00388.html Allow to use more than one page for a ring. • Still need to increase BLKIF_MAX_SEGMENTS_PER_REQUEST to support more segments in one request • Request structure size gets increased as a result Move seg[BLKIF_MAX_SEGMENTS_PER_REQUEST] out of the request structure and put them in a separate ring instead. • More scalable 14
  • 15. Result of Multi-page-ring Source: Intel Multi-page-ring improved the peak bandwidth by 17% and reduced the overhead by 29% at the same time. • Still a big gap with the bare-metal. Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 15 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 16. Per-device-lock Again Source: Intel With multi-page-ring, per-device-lock can help get even higher b/w and lower overhead. Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 16 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 17. Review the Changes Source: Intel Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those 17 factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance.
  • 18. Summary HVM outperforms PV guest in both CPU/memory intensive workload and disk I/O intensive workload. Xen PV disk performance is 26% worse than bare-metal in terms of peak bandwidth. We proposed a multi-page-ring patch that improves the PV disk performance by 24% and reduces the gap with bare-metal to single digit. 18
  • 19. Q&A from the Xen Summit Q: There is another queue, the queue of the outstanding requests, did you increase the request queue? A: Yes, the data shown in here was with increased outstanding I/O numbers which are large enough to reach peak b/w. Q: On the PV vs. HVM work, what was the kernel that was being use for SPCjbb, and what distribution? A: see backup pages for detailed configurations. Q: How many guests where running? A: single guest VM. Q: Do you have inside into how much the hypervisor contributes to the difference between hypercall and vmexit? A: Hypervisor should have no direct contribution to the difference. The difference is majorly from the different guest kernels. Q: How much of these differences might be helped with APICv? A: In this particular case (SPECjbb), APICv won’t help much because SPECjbb is CPU/memory intensive and has few external interrupts. APICv may help a lot in an I/O intensive workload. What processor was this done on? A: see backup pages for detailed configurations. Q: How was the data (patch lengh, CPI, cache miss rate) collected? A: using Intel in-house tool. Q: Was this done with ept? A: Yes Q: What about shadow? A: Did not try it on shadow. Shadow and PV should have similar performance because the use the same page table format. Q: Use huge pages with specjbb will make a big difference, because it reduces the number of page walks greatly. A: guest hugepage was disabled to get a fair comparison between PV and HVM. See backup for configurations. Q: Was the test run on rotating media or SSD? A: 19 all on SSD.
  • 21. Configurations - SPECjbb • H/W: 2-socket Intel® Xeon® Processor X5660 with 48GB memory (24GB on each node) running at 800MHz. • EIST/Turbo disabled; core c3/c6 disabled. • VMM: Xen 4.1.0 • Dom0: Linux 2.6.32 with 2 vCPUs and 4GB memory • Guest O/S: RHEL 6.1 • THP is disabled for a fair comparison, because PV guest does not support huge page. • Implies 4KB guest page since we did not reserve huge page explicitly. • With RHEL6.1’s PVOPS kernel, PV disk and PV network are used for HVM • It does not matter much for this particular workload (SPECjbb has almost zero disk/network activities) • VM Configurations: • Case 1: 4 vCPUs + 8GB memory • pinned to 4 pCPUs as a group (floating) • the 4 pCPUs share 2 cores (HT enabled) in socket 0 • Case 2: 12 vCPUs + 19GB memory • pinned to 12 pCPUs as a group (floating) • the 12 pCPUs share the socket 0 (HT enabled) • Workload Configurations: • jdk1.6.0_27/bin/java -server –Xmx<N>g –Xms<N>g -XX:+UseParallelOldGC • Case 1: 6GB heap size; Case 2: 18GB heap size • Workload methodology: • Case 1: 3 warehouses for ramp-up and 4 warehouses for steady state. • Case 2: 11 warehouses for ramp-up and 12 warehouses for steady state. • 10 minutes ramp-up, and 15 minutes steady state. • Used 60 minutes steady state for data collection. 21
  • 22. Configurations - SysBench H/W: 2-socket Intel® Xeon® Processor E5-2680 with 32GB DDR3 memory running at 1333MHz. 1x Intel SSD for MySQL data. VMM: Xen 4.1.2 • Dom0: Linux 3.2.9 Guest O/S: RHEL 6.2 • With RHEL6.2’s PVOPS kernel, PV disk and PV network were used in both HVM mode and PV guest mode. VM Configurations: • #vCPUs = #pCPUs; 4GB memory • The data disk was assigned to VM in ‘PHY:’ mode. Workload Configurations: • SysBench: prepared 40 million rows ~9.2GB data • MySQL: set innodb buffer pool size to 2GB and use O_DIRECT as flush mode. • SysBench Read-only test: sysbench --mysql-db=sbtest --max-requests=0 --test=oltp -- mysql-table-engine=innodb --oltp-table-size=40000000 --mysql-user=root --mysql- host=localhost --mysql-socket=/var/lib/mysql/mysql.sock --mysql-password=123456 -- num-threads=$vCPU_num --max-time=300 --oltp-read-only=on run 22
  • 23. Configurations – PV Disk Hardware: • 2-socket Intel® Xeon® Processor X5660 with 48GB memory • HBA: LSI 9205 • Storage: 2 disk bays with 4 Intel SSD disks in each. Software: • Hypervisor: Xen 4.1.2 • Dom0: Linux 3.3.6 (8 vCPUs pinned to 8 pCPUs) • Guest O/S: Linux 3.3.6 (8 vCPUs pinned to 8 pCPUs; 2GB memory) • Virtual disks are configured as ‘PHY:’, i.e., no caching in dom0 • Workload: aio-stress • Accessing disks in O_DIRECT mode, i.e, no caching in guest. • Using 64KB sequential I/O pattern 23