[241] 하나의 cpu 에 운영제체 두 개 김성민

1. 아무 짝에도 쓸모 없을 것 같던
하나의 CPU 에 운영체제
두 개 부팅시키기
김성민
㈜구름네트웍스
1

2. AUTHORS
2
김성호
sungho@gurum.cc
양유석
youseok@gurum.cc

3. CONTENTS
1. Hypervisor, Container, 그리고 Multi-Kernel Architecture
2. 이걸 도대체 어디에 쓰나?
3. PopcornLinux 분석
4. Linux 2개 부팅 시키기
5. Linux + PacketNgin 부팅 시키기
6. Linux + PacketNgin Network Performance
3

4. 1.
Hypervisor, Container,
그리고
Multi-Kernel Architecture
4

5. 1.1 Hypervisor
5
Source: vmware.com
Type 1 Hypervisor Type 2 Hypervisor

6. 1.1 Hypervisor
6
Source: vmware.com
Type 1 Hypervisor Type 2 Hypervisor

7. 1.1 Hypervisor
7
VMware vSphere/ESXi
Microsoft Hyper-V
Citrix Xen
Linux KVM (??)
VMware Workstation
Oracle VirtualBox
QEMU
Linux KVM (??)
Source: searchservervirtualization.techtarget.com/news/2240034817/
KVM-reignites-Type-1-vs-Type-2-hypervisor-debate

8. 1.2 Container
8Source: cloudacademy.com/blog/container-virtualization/

9. 1.2 Container
9Source: cloudacademy.com/blog/container-virtualization/

10. 1.2 Container
10
Linux Docker
Linux LXC
Linux OpenVZ
Solaris Container
FreeBSD jail
Source: cloudacademy.com/blog/container-virtualization/

11. 1.3 Multi-Kernel Architecture
11Source: popcornlinux.org

12. 1.3 Multi-Kernel Architecture
12Source: popcornlinux.org

13. 1.3 Multi-Kernel Architecture
13
Popcorn Linux – Linux + Linux
Nucleus RTOS – Linux + Nucleus
PacketNgin RTOS – Linux + PacketNgin
Network Processors – Linux + Firmware
Source: popcornlinux.org

14. 2.
이걸 도대체 어디에 쓰나?
14

15. 2.1 모바일 분야 – ARM TrustZone
15Source: www.arm.com/products/processors/technologies/trustzone/tee-smc.php

16. 2.1 모바일 분야 – ARM TrustZone
16Source: www.arm.com/files/pdf/TrustZone-and-FIDO-white-paper.pdf

17. 2.2 서버 분야 – Many-core Architecture
Source: blogs.intel.com/technology/2012/06/intel-xeon-phi-coprocessors-accelerate-discovery-and-innovation/ 17

18. 2.2 서버 분야 – Many-core Architecture
18Source: www.ssrg.ece.vt.edu/theses/MS_Katz.pdf

19. 2.2 서버 분야 – Many-core Architecture
19Source: www.pugetsystems.com/labs/articles/Adobe-After-Effects-CC-2015-Multi-Core-Performance-714/

20. 2.2 서버 분야 – Many-core Architecture
20Source: queue.acm.org/detail.cfm?id=2000516

21. 2.2 서버 분야 – Many-core Architecture
21Source: www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html

22. 2.3 네트워크 분야 –Control Plane/Data Plane
22Source: www.linkedin.com/pulse/6wind-from-data-plane-acceleration-virtual-appliances-humair-ahmed

23. 3.
PopcornLinux 분석
23Source: http://www.popcornlinux.org/sources/slide/VTLUUG_Tech_may2013.pdf

24. BIOS
Disk의 MBR을 읽어 메모리에 로딩
3.1 O/S 부팅 과정 (x86_64, BIOS 기준)
24
BSP(BootStrap Processor), Core 0
AP(Application Processor), Core 1 ~ n
MBR (Stage 1, 16bit)
Stage 2 로더를 메모리에 로딩
Stage 2 로더 (16bit)
Kernel을 메모리에 로딩
16bit -> 32bit 전환
Kernel (32bit)
커널 재배치
32bit -> 64bit 전환
로더 (trampoline_64.S, 16bit)
Kernel을 메모리에 로딩
16bit -> 32bit 전환
Kernel (32bit)
커널 재배치
32bit -> 64bit 전환

25. 3.2 Linux Kernel Booting
25Source: www.linux-arm.org/LinuxBootLoader/SMPBoot

26. 3.3 PopcornLinux Booting
26
CPU Memory PCI
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
0M
~
8GB
NIC0 NIC1 NIC2 NIC3

27. 3.3 PopcornLinux Booting (First Booting)
27
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU Memory
0M
~
8GB
NIC0
PCI
NIC1 NIC2 NIC3
PopcornLinux Kernel

28. 28
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU Memory
0M
~
8GB
NIC0
PCI
NIC1 NIC2 NIC3
PopcornLinux Kernel
1. Core의 개수
2. Memory의 용량
3. PCI 디바이스 목록
4. CPU, Memory, PCI 할당 규칙
3.3 PopcornLinux Booting (First Booting)

29. 3.3 PopcornLinux Booting (Master Booting)
29
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU Memory
0M
~
8GB
NIC0
PCI
NIC1 NIC2 NIC3
PopcornLinux Master Kernel

30. 3.3 PopcornLinux Booting (Secondary Booting)
30
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU Memory
0M
~
8GB
NIC0
PCI
NIC1 NIC2 NIC3
PopcornLinux Master Kernel
PopcornLinux Secondary Kernel

31. 3.3 PopcornLinux Booting (Source Code)
31
Booting /arch/x86/include/asm/bootparam.h
Booting /arch/x86/kernel/e820.c
Booting /arch/x86/kernel/head64.c
Booting /arch/x86/kernel/process_64.c
Booting /arch/x86/kernel/setup.c
Booting /arch/x86/kernel/smpboot.c
Booting /arch/x86/kernel/trampoline.c
Booting /arch/x86/kernel/trampoline_64_bsp.S
Booting /arch/x86/kernel/vmlinux.lds.S
Booting /include/linux/multikernel.h
Booting /kernel/multikernel.c

32. 3.3 PopcornLinux Booting (Source Code)
32
CPU /arch/x86/include/asm/apic.h
CPU /arch/x86/kernel/apic/io_apic.c
CPU /arch/x86/kernel/cpu/common.c
CPU /arch/x86/kernel/i387.c
CPU /arch/x86/kernel/traps.c
CPU /drivers/acpi/processor_driver.c
CPU /include/linux/sched.h
Memory /arch/x86/mm/dump_pagetables.c
Memory /mm/memblock.c
Memory /mm/nobootmem.c

33. 3.3 PopcornLinux Booting (Source Code)
33
Interrupt /arch/x86/include/asm/unistd_64.h
Interrupt /include/linux/syscalls.h
VTY /arch/x86/include/asm/entry_arch.h
VTY /drivers/tty/serial/8250.c
VTY /drivers/tty/tty_buffer.c
VTY /drivers/tty/vty.c

34. Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
3.4 CPU Partitioning (Overview)
34
CPU
1. Booting parameter 추가
2. Linux Kernel의 CPU mask 정보 조작
linux/arch/x86/setup.c
arch/x86/kernel/smpboot.c

35. 3.4 CPU Partitioning (Master Booting)
35
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU
1. Core 0번에서 Kernel 부팅
1. BIOS가 MBR(Stage 1 Loader) 로딩 (16bit mode)
2. Stage 1 Loader는 Stage 2 Loader 로딩 (16bit mode)
3. Stage 2 Loader는 Kernel 로딩 (16bit -> 32bit mode)
4. Kernel은 메모리 주소 재배치 (32bit -> 64bit mode)

36. 3.4 CPU Partitioning (Master Booting)
36
1. INIT IPI (Inter-Process Interrupt)
2. 100us wait
3. INIT IPI
4. 100us wait
5. STARTUP IPI
6. 200us wait
INIT, INIT, STARTUP IPI
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU 1. Core 0번에서 Kernel 부팅
2. Core 0번이 Core 1~3에 Startup Interrupt 전송
(smpboot.c)

37. 3.4 CPU Partitioning (Master Booting)
37
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU
1. Core 0번에서 Kernel 부팅
2. Core 0번이 Core 1~3에 Startup Interrupt 전송
(smpboot.c)
3. Core 1~3번은 Kernel 부팅 (trampoline_64.S)
1. Core 1~3은 trampoine_64.S 를 실행
2. trampoine_64.S는 CPU mode를
16bit -> 32bit로 전환
3. Kernel로 Jump
4. Kernel은 32bit -> 64bit 전환

38. 3.4 CPU Partitioning (Secondary Booting)
38
1. Primary Linux은
Secondary Linux의 BSP (BootStrap
Processor) Kernel을 부팅
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU
INIT, INIT, STARTUP IPI
1. INIT IPI (Inter-Process Interrupt)
2. 100us wait
3. INIT IPI
4. 100us wait
5. STARTUP IPI
6. 200us wait

39. 3.4 CPU Partitioning (Secondary Booting)
39
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU 1. Primary Linux은
Secondary Linux의 BSP (BootStrap
Processor) Kernel을 부팅
2. Secondary Linux의 BSP Kernel 부팅
1. BIOS는 trampoline_64_bsp.S를 실행
2. trampoline_64_bsp.S는 Kernel 로딩
(16bit -> 32bit mode)
3. Kernel은 메모리 주소 재배치
(32bit -> 64bit mode)

40. 3.4 CPU Partitioning (Secondary Booting)
40
INIT, INIT, STARTUP IPI
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU 1. Primary Linux은
Secondary Linux의 BSP (BootStrap
Processor) Kernel을 부팅
2. Secondary Linux의 BSP Kernel 부팅
3. BSP Kernel은 AP Kernel 부팅

41. 3.4 CPU Partitioning (Secondary Booting)
41
Core 0 Core 1 Core 2 Core 3
Core 4 Core 5 Core 6 Core 7
CPU 1. Primary Linux은 Secondary Linux의 BSP
(BootStrap Processor) Kernel을 부팅
2. Secondary Linux의 BSP Kernel 부팅
3. BSP Kernel은 AP Kernel 부팅
4. Secondary Linux의 AP Kernel 부팅 완료
1. Core 1~3은 trampoine_64.S 를 실행
2. trampoine_64.S는 CPU mode를
16bit -> 32bit로 전환
3. Kernel로 Jump
4. Kernel은 32bit -> 64bit 전환

42. 3.5 Memory Partitioning (Overview)
42
1. Booting parameter 추가
2. Page table 조작
Memory
0M
~
8GB
Primary Linux Secondary Linux
0MB 4GB 0MB 4GB

43. 43Source: cse-wiki.unl.edu/wiki/index.php/Computer_Architecture_Notes
3.5 Memory Partitioning (Overview)

44. 44

45. 3.5 Memory Partitioning (Overview)
45
1. Booting parameter 추가
2. Page table 조작
linux/arch/x86/kernel/e820.c
linux/arch/x86/kernel/e820.c
Memory
0M
~
8GB
Primary Linux Secondary Linux
0MB 4GB 0MB 4GB

46. 3.5 Memory Partitioning (Master Booting)
46
Memory
0M
~
8GB
1. Booting parameter 추가
memmap=4G@0M mem=4G
1. e820.c에서 0~4GB 영역을 사용 가능한 영역으로 포함 시킴
2. 메모리는 최대한 4GB만 할당함

47. 3.5 Memory Partitioning (Master Booting)
47
1. Booting parameter 추가
memmap=4G@0M mem=4G
2. Page Table 설정
arch/x86/mm/init_64.c
Memory
0M
~
8GB

48. Memory
0M
~
8GB
1. e820.c에서 0~4GB 영역을 사용
불가능한 영역으로 포함 시킴
2. 메모리는 최대한 8GB만 할당함
(8GB – 4GB = 4GB)
3.5 Memory Partitioning (Secondary Booting)
48
1. Booting parameter 추가
memmap=4G@0M mem=4G
2. Page Table 설정
3. Booting parameter 추가
memmap=4G$0M mem=8G

49. 3.5 Memory Partitioning (Secondary Booting)
49
1. Booting parameter 추가
memmap=4G@0M mem=4G
2. Page Table 설정
3. Booting parameter 추가
memmap=4G$0M mem=4G
4. Page Table 설정
arch/x86/mm/init_64.c
Memory
0M
~
8GB

50. 3.6 PCI Device Partitioning (Overview)
50
1. PCI black list를 Kernel 파라미터로 받음
2. 사용 가능한 PCI 목록에서 black list 제외
PCI 0 PCI 2 PCI 3PCI 1

51. 3.6 PCI Device Partitioning (Overview)
51
Bus, Slot, Function Vendor ID, Device ID

52. 3.6 PCI Device Partitioning (Master Booting)
52
1. PCI black list를 Kernel 파라미터로 받음
pci_dev_flag=0x8086:0x7010:b,0x8086:0
x100e:b
linux/drivers/pci/probe.c
PCI 2 PCI 3PCI 0 PCI 1
1. Boot 파라미터 해석
2. Black list에 해당 PCI device 포함
3. PCI probe 단계에서 해당 PCI
device는 제외

53. 53

54. 3.6 PCI Device Partitioning (Secondary Booting)
54
1. PCI black list를 Kernel 파라미터로 받음
2. 사용 가능한 PCI 목록에서 black list 제외
3. Master Kernel에서 사용하는 PCI 목록을
black list에 파라미터로 받음
4. 사용 가능한 PCI 목록에서 black list 제외
PCI 0 PCI 2 PCI 3PCI 1
1. Boot 파라미터 해석
2. Black list에 해당 PCI device 포함
3. PCI probe 단계에서 해당 PCI
device는 제외

55. 3.7 Multi-Kernel 간의 통신
55
Memory
0M
~
8GB
1. Kernel 간에 공유하는 메모리 공간 할당
2. 모든 Kernel에게 해당 공간 주소를 Kernel 파라
미터로 넘겨줌
3. Kernel들은 공유 메모리를 통해 정보를 주고 받음
linux/drivers/tty/vty.c
VTY , Tunnel 영역

56. 4.
Linux 2개 부팅 시키기
PopcornLinux 버그 수정 버전
https://github.com/packetngin/popcorn
56

57. 4.1 Linux 설치
57

58. 4.2 유틸리티 설치
58

59. 4.3 Patch Download
59

60. 4.4 Linux Kernel Download
60

61. 4.5 Patch Linux Kernel
61

62. 4.6 Build Kernel
62

63. 4.7 Kexec Download
63

64. 4.8 Patch Kexec
64

65. 4.9 Build Popcorn Utils
65

66. 4.10 VTY Configuration
66
Boot parameter 설정
Grub 설정

67. 4.11 Reboot
67

68. 4.12 Master Linux Booting
68
Core가 1개만 할당된
것을 확인

69. 4.12 Master Linux Booting
69
메모리가 약 870MB 할당된 것을 확인
(VTY 등 영역 제외 한 수치)

70. 4.13 Build Secondary Linux Image
70
ELF 형식에서 .notes, .comment 섹션을 삭제

71. 4.14 Booting Secondary Linux
71
1. RAM disk 복사
copy_ramdisk $RAMDISK_ADDR $RAMDISK_FILE
2. Kernel parameter 복사
wr_cmdline $BOOT_ARGS
3. Kernel image 복사
kexec -d -a $BOOT_ADDR -l $KERNEL -t elf-x86_64 --args-none
4. Kernel 부팅
kexec -d -a $BOOT_ADDR -b $CPU

72. 4.15 Secondary Linux 접속
72

73. 4.16 Secondary Linux 확인
73
1번 Core 1개만 할당된
것을 확인

74. 5.
Linux + PacketNgin 부팅 시키기
74

75. 5.1 PacketNgin RTOS 소개 (1/4)
75
Eth IP TCP Payload Ether Block
IP Block
TCP Block
Web
Browser
Kernel Space
User Space
NICEth IP TCP Payload
IP TCP Payload
TCP Payload
Payload
General Purpose O/S
Eth IP TCP Payload Ether Block
Firewall
Kernel Space
User Space
NICEth IP TCP Payload
Eth IP TCP Payload
Network O/S

76. 5.1 PacketNgin RTOS 소개 (2/4)
76
Programmability
• Linux는 Host Network Programming 하기에 적합한 O/S
• PacketNgin은 Network Node Programming 하기에 적합한 O/S
• ARP, ICMP, IPsec 소스 코드의 양이 Linux에 비해 2/3 ~ 1/2 수준

77. 5.1 PacketNgin RTOS 소개 (3/4)
77
+ Network H/W depedent code
+ deliver_skb()
+ ret = pt_prev->func(skb, skb->dev, pt_prev);
+ ip_rcv()
+ nf_hook()
+ ip_rcv_finish()
+ ip_route_input()
+ dst_input()->ip_forward() or ip_input()
+ ip_input // Remove the IPv4 header
+ ip_input_finish
+ ret = ipprot->handler(&skb, &nhoff);
+ xfrm4_rcv()
+ xfrm_input()
+ xfrm4_parse_spi()
+ xfrm_state_lookup() // lookup IPsec SA
+ xfrm_beet_input(skb, x) //To change to inner IP header.
+ nexthdr = x->type->input(x, xfrm.decap, skb) // ==
esp_input
+ esp_input() // process ESP based on inner
address
+ returns 0 ;
+ /* beet handling in xfrm_rcv_spi */
+ netif_rx()
+ // ip_input_finish returns 0
+ // netif_receive_skb returns 0
+netif_receive_skb // Now we have an IPv4 packet. So the input flow is
for v4 packet.
+ deliver_skb()
+ ret = pt_prev->func(skb, skb->dev, pt_prev);
+ ip_rcv()
+ nf_hook() //This calls ip_rcv_finish(skb)
+ ip_rcv_finish() // Here the skb->dst is NULL and so is filled for
the input side.
+ ip6_route_input()
+ dst_input()->ip_forward() or ip_input()
+ ip_input // Remove the IPv4 header
+ ip_input_finish
+ …
+ Network H/W depedent code
+ nic_process_output()
+ fifo_push()
+ ni_input()
+ ipsec_inbound()
+ sad_get()
+ ipsec_decrypt()
+ spd_get()
+ ni_output()

78. 5.1 PacketNgin RTOS 소개 (4/4)
78

79. 5.2 Porting – Popcorn Utility
79
kexec 변경
1. Linux는 16MB 영역에 로딩
2. PacketNgin은 4MB 영역에 로딩
3. Hard-coding address => 16MB
Booting utility
1. Hard-coding address => 16MB
Hard-coding address => 4MB
Hard-coding address => 4MB

80. 5.3 Porting – Kernel 변경
80
Booting logic 변경
1. BSP Booting – Grub과 호환
2. AP Booting – 자체 제작 loader에서
16bit -> 32bit 전환
Standard I/O 방식 변경
1. VGA Buffer에 직접 출력하는 방식
2. 자체 제작한 VGA 드라이버 사용
1. BSP Booting 필요 없음
2. Kexec와 호환되도록 loader 수정
1. Master Linux Kernel과 충돌
2. Serial 통신으로 대체
3. (차후)VTY와 호환 되도록 수정

81. 5.3 Porting – Kernel 변경
81
Memory map 변경
1. PacketNgin Kernel은 2~6MB 사용
Kernel parameter 전달 방식 변경
1. Grub2에서 제공되는
Multiboot2 스펙
1. Linux Kernel과 PacketNgin
Kernel 영역이 겹치는 문제
2. PacketNgin Kernel을 4~8MB로
이동
1. Linux boot parameter 스펙과
호환되도록 변경

82. 5.3 Porting – Kernel 변경
82
Linux Inter-Processor Interrupt와 호환되도록 변경
1. Linux IPI 번호 체계와
PacketNgin IP 번호 체계 충돌
1. Linux에서 사용하지 않는 IPI 번호를
PacketNgin IPI 번호로 할당함

83. 5.3 Porting – Manager App 변경
83
PacketNgin Manager -> Linux로 포팅
1. PacketNgin Shell
2. PacketNgin Manage API
3. VM 관리 API
4. App 관리 API
1. Linux Shell로 대체
2. Linux에서 동작하는 Manage API
3. Linux에서 동작하는 VM 관리 API
4. Linux에서 동작하는 App 관리 API

84. 5.3 Porting – Network I/O 고속화
84
Interrupt 방식 -> Polling 방식 변경
1. Polling 방식으로 고속으로 I/O
2. PacketNgin 자체
Device Driver 사용
1. Linux의 NAPI 최대한 활용
2. Interrupt와 Polling 방식 병행
3. Linux Device Driver 최소한으로
수정
Packet Processing API 변경
1. PacketNgin 자체 Packet 구조체
1. sk_buff <-> Packet 구조체 간
변환 로직 구현

85. 6.
Linux + PacketNgin
Network Performance
85

86. 6.1 Goal – 6WindGate Performance
86Source: www.slideshare.net/openstack_kr/6-wind-openstackkoreameetup27may15/4

87. 6.1 Goal – 6WindGate Performance
87Source: blog.ipspace.net/2012/02/6wind-solving-virtual-appliance.html

88. 6.1 Goal – 6WindGate Performance
88Source: www.linkedin.com/pulse/6wind-from-data-plane-acceleration-virtual-appliances-humair-ahmed

89. 6.2 Goal – PacketNgin 1.0
89
Core i5 + NetFPGA NIC +420%

90. Linux
6.3 Network I/O– Linux vs PacketNgin
90
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
UDP
Echo
Eth IP UDP Payload Ether Block
IP Block
UDP Block
UDP
Echo
Kernel Space
User Space
NICEth IP UDP Payload
IP UDP Payload
UDP Payload
Payload

91. 6.3 Network I/O– Linux vs PacketNgin
91
UDP
Echo
Eth IP UDP Payload Ether Block
UDP Echo
Kernel Space
User Space
NICEth IP UDP Payload
Eth IP UDP Payload
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNgin

92. 6.3 Network I/O–PacketNgin 2.0
92
UDP
Echo
Packet
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNgin
1. Packet이 NIC에 도착

93. 6.3 Network I/O–PacketNgin 2.0
93
1. Packet이 NIC에 도착
2. Linux의 NIC Driver에서 Packet을 확인
UDP
Echo
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNginPacket

94. 6.3 Network I/O–PacketNgin 2.0
94
1. Packet이 NIC에 도착
2. Linux의 NIC Driver에서 Packet을 확인
3. NIC Driver에서 패킷을 복제해
PacketNgin Kernel로 전달
Packet
UDP
Echo
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNgin

95. 6.3 Network I/O–PacketNgin 2.0
95
Packet 1. Packet이 NIC에 도착
2. Linux의 NIC Driver에서 Packet을 확인
3. NIC Driver에서 패킷을 복제해
PacketNgin Kernel로 전달
4. PacketNgin Kernel은 UDP Echo App
에게 Packet을 전달
5. UDP Echo App은 Packet을 처리
UDP
Echo
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNgin

96. 6.3 Network I/O–PacketNgin 2.0
96
1. Packet이 NIC에 도착
2. Linux의 NIC Driver에서 Packet을 확인
3. NIC Driver에서 패킷을 복제해
PacketNgin Kernel로 전달
4. PacketNgin Kernel은 UDP Echo App
에게 Packet을 전달
5. UDP Echo App은 Packet을 처리
6. PacketNgin Kernel -> Linux Kernel -
> NIC 순서로 Packet 출력
UDP
Echo
NIC
Core 0 Core 1 Core 2 Core 3
0M
~
8GB
Linux PacketNgin
Packet

97. 6.4 Result–UDP Echo Performance
97

98. 6.5 Wrap-Up
98
1. Multi-Kernel Architecture
2. Popcorn Linux
3. Control Plane / Data Plane
4. Performance – 130~1,300%↑

99. Q & A
99
PopcornLinux 버그 수정 버전
https://github.com/packetngin/popcorn
PacketNgin 2.0 Preview
https://goo.gl/rxET1T

100. Thank You
100