SpeQuloS: A QoS Service for BoT Applications Using Best Effort Distributed Computing Infrastructures
Simon Delamare Gilles Fedak Derrick Kondo Oleg Lodygensky
High-Performance Parallel and Distributed Computing, 2012
SpeQuloS: A QoS Service for BoT Applications Using Best Effort Distributed Computing Infrastructures
1. SpeQuloS: A QoS Service for BoT Applications Using
Best Effort Distributed Computing Infrastructures
Simon Delamare 1
Gilles Fedak 2
Derrick Kondo 3
Oleg Lodygensky 4
1
LIP/CNRS, Univ. Lyon, France
2
LIP/INRIA, Univ. Lyon, France
3
LIG/INRIA, Univ. Grenoble, France
4
LAL/CNRS, Univ. Paris XI, France
High-Performance Parallel and Distributed Computing, 2012
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 1 / 18
2. Introduction
BE-DCI = “Best-Effort” Distributed Computing Infrastructure
→ Large computing power at low cost, Avoid wasting resources
→ No availability guarantee
Desktop Grids
→ BOINC projects: Peta FLOPS for free
Grids used in Best-Effort mode
→ ≈ 40% of utilization in Grid5000@Lyon
Cloud “Spot” Instances
→ c1.large instance price: 0.12$/h (spot) vs. 0.32$/h (regular)
Relevant for BoT execution ...
Bag of Tasks: Set of independent tasks to compute
→ but Low QoS level
Especially compared to regular infrastructures
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 2 / 18
3. Performance Problem Addressed
BoT completion rate increases at the end of execution
→ Tail Effect
0
0.2
0.4
0.6
0.8
1
1.2
0 20 40 60 80 100
BoTcompletionratio
Time
Continuation is performed
at 90% of completion
Ideal Time Actual Completion Time
Tail Duration
Slowdown = (Tail Duration + Ideal Time) / Ideal Time
BoT completion
Tail part of the BoT
Measured by Slowdown:
S =
IdealCompletionTime
RealCompletionTime
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 3 / 18
4. Slowdown by Tail Effect
Slowdown reported on BoT execution
0
0.2
0.4
0.6
0.8
1
0.1 1 10 100
Fractionofexecutionwheretailslowdown<S
Tail Slowdown S (Completion time observed divided by ideal completion time)
BOINC
XWHEP
Best 50% ⇒ S < 1.3
25% to 33% ⇒ S > 2
Worst 5% ⇒ S> 4 to 10
Avg. % of BoT in tail Avg. % of time in tail
BE-DCI Trace BOINC XWHEP BOINC XWHEP
Desktop Grids 4.65 5.11 51.8 45.2
Best Effort Grids 3.74 6.40 27.4 16.5
Spot Instances 2.94 5.19 22.7 21.6
→ Caused by no more than the last 7% of
BoT
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 4 / 18
5. How to improve the situation ?
Better scheduling
QoS in Grid scheduling ([12], [20], [38])
→ Require heavy modification of middleware
→ No satisfactory solution for unreliable infrastructure ([7])
Addressing the tail effect
→ e.g. in MapReduce ([3], [39]), but require precise information from compute
nodes, hard in large DCIs.
Building Hybrid DCIs
Grid & Desktop Grid ([35],[36])
→ Mostly to offload Grid usage
Using Cloud computing ([10],[28],[37])
→ To address peak demands
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 5 / 18
6. SpeQuloS Service
→ Improving BE-DCIs users perceived QoS
Speeding up BoT execution
Bring information on expected BoT execution time
By dynamic provision of Cloud resources
→ Monitoring BoT execution
→ Execute the tail on Cloud
Features:
1 Our context: Existing BE-DCIs and Clouds, not administrator: Black Boxes
2 Interface with users: QoS requests, State of completion, Prediction on
remaining time
3 Careful utilization of Cloud resources w/ Billing & Accounting of usage
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 6 / 18
7. Framework
SpeQuloS modules:
Information: Collect QoS-related
information from DGs
Oracle: Strategies to appropriately
use Cloud resources / QoS
prediction for users
Scheduler: Start/Stop Cloud
resources, usage accounting
Credit System: Bill Cloud usage to
user, using “credits” to buy Cloud
resource cpu.h
Implementation
Independant modules using Python & MySQL
Supported Clouds: EC2, OpenNebula, etc.
Supported DG middleware: BOINC & XtremWeb-HEP
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 7 / 18
8. Cloud Provisioning Strategies
When to start Cloud resources ?
At 90% of BoT completion (9C)
At 90% of BoT assignment (9A)
When Tail appear, by monitoring execution time variance (V)
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 8 / 18
9. Cloud Provisioning Strategies
When to start Cloud resources ?
At 90% of BoT completion (9C)
At 90% of BoT assignment (9A)
When Tail appear, by monitoring execution time variance (V)
How many Cloud resources to start (for a given amount of Credits) ?
Greedy: As much as possible, for 1 hour of cloud usage (G)
Conservative: To ensure that there will be enough credits to run Cloud up to
an estimated completion time (C)
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 8 / 18
10. Cloud Provisioning Strategies
When to start Cloud resources ?
At 90% of BoT completion (9C)
At 90% of BoT assignment (9A)
When Tail appear, by monitoring execution time variance (V)
How many Cloud resources to start (for a given amount of Credits) ?
Greedy: As much as possible, for 1 hour of cloud usage (G)
Conservative: To ensure that there will be enough credits to run Cloud up to
an estimated completion time (C)
How to use Cloud resources ?
Flat: Cloud worker not differentiated from BE-DCI workers (F)
Reschedule : Scheduler reshedule tasks executed on BE-DCI to Cloud (R)
Cloud Duplication : Uncompleted tasks are duplicated to a dedicated Cloud
infrastructure (D)
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 8 / 18
11. Experimentation Setup (1)
Simulations using real BE-DCI infrastructures availability traces, various BoT
workloads, BOINC and XWEP middleware
BE-DCIs availability traces :
Desktop Grids: seti, nd (SETI@Home & NotreDame traces from FTA)
Best Effort Grids: g5klyo, g5kgre (Available ressources in Grid5000 Lyon &
Grenoble clusters in December 2010)
Cloud Spot instances: spot10, spot100 (Maximum number of instances for a
renting cost of 10 or 100 $ per hour, fluctuates according to market price)
trace length mean deviation min max av. quartiles (s) unav. quartiles (s) avg. power power
(days) (nops/s) std. dev.
seti 120 24391 6793 15868 31092 61,531,5407 174,501,3078 1000 250
nd 413.87 180 4.129 77 501 952,3840,26562 640,960,1920 1000 250
g5klyo 31 90.573 105.4 6 226 21,51,63 191,236,480 3000 0
g5kgre 31 474.69 178.7 184 591 5,182,11268 23,547,6891 3000 0
spot10 90 82.186 3.814 29 87 4415,5432,17109 4162,5034,9976 3000 300
spot100 90 823.95 4.945 196 877 1063,5566,22490 383,1906,10274 3000 300
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 9 / 18
12. Experimentation Setup (2)
BoT workloads:
Size nops / task Arrival time
SMALL 1000 3600000 0
BIG 10000 60000 0
RANDOM norm(µ = 1000, σ2
= 200) norm(µ = 60000, σ2
= 10000) weib(λ = 91.98, k = 0.57)
Simulations methodology:
Reproducible executions wo & w/ SpeQuloS
SpeQuloS Credits provisioned w/ 10% of BoT workload (in Cloud resource
cpu.hour equivalent)
→ 25000 BoT execution traces
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 10 / 18
13. Strategies Comparison
Tail Removal Efficiency
→ Tail Duration w/ SpeQuloS vs Tail Duration wo SpeQuloS
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P
Tail Removal Efficiency (Percentage P)
9C-G-F
9A-G-F
V-G-F
9C-C-F
9A-C-F
V-C-F
Flat deployment
strategy
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P Tail Removal Efficiency (Percentage P)
9C-G-R
9A-G-R
V-G-R
9C-C-R
9A-C-R
V-C-R
Reschedule deployment
strategy
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P
Tail Removal Efficiency (Percentage P)
9C-G-D
9A-G-D
V-G-D
9C-C-D
9A-C-D
V-C-D
Cloud duplication
deployment strategy
Best strategies are able to
Suppress tail for 50% of execution
Half the tail for 80% of execution
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 11 / 18
14. Strategies Comparison
Tail Removal Efficiency
→ Tail Duration w/ SpeQuloS vs Tail Duration wo SpeQuloS
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P
Tail Removal Efficiency (Percentage P)
9C-G-F
9A-G-F
V-G-F
9C-C-F
9A-C-F
V-C-F
Flat deployment
strategy
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P Tail Removal Efficiency (Percentage P)
9C-G-R
9A-G-R
V-G-R
9C-C-R
9A-C-R
V-C-R
Reschedule deployment
strategy
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100
FractionofBoTwheretailefficiency>P
Tail Removal Efficiency (Percentage P)
9C-G-D
9A-G-D
V-G-D
9C-C-D
9A-C-D
V-C-D
Cloud duplication
deployment strategy
Best strategies are able to
Suppress tail for 50% of execution
Half the tail for 80% of execution
Flat (F) < Reschedule (R) & Cloud Duplication (D)
Tail Detection (V) triggers Cloud too late
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 11 / 18
15. Cloud Resources Consumption
Percentage of credits spent vs
credits provisioned (=10% of BoT
workload).
10% to 25% of what has been
provisioned are actually used by
Cloud resources
0
10
20
30
40
50
9C-G
-F
9C-G
-R
9C-G
-D9C-C-F
9C-C-R
9C-C-D9A
-G
-F
9A
-G
-R
9A
-G
-D9A
-C-F
9A
-C-R
9A
-C-DV
-G
-F
V
-G
-R
V
-G
-DV
-C-F
V
-C-R
V
-C-D
Percentageofcreditsused
Combination of SpeQuloS strategies
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 12 / 18
16. Cloud Resources Consumption
Percentage of credits spent vs
credits provisioned (=10% of BoT
workload).
10% to 25% of what has been
provisioned are actually used by
Cloud resources
0
10
20
30
40
50
9C-G
-F
9C-G
-R
9C-G
-D9C-C-F
9C-C-R
9C-C-D9A
-G
-F
9A
-G
-R
9A
-G
-D9A
-C-F
9A
-C-R
9A
-C-DV
-G
-F
V
-G
-R
V
-G
-DV
-C-F
V
-C-R
V
-C-D
Percentageofcreditsused
Combination of SpeQuloS strategies
→ ≈2.5% of BoT workload is executed on Cloud
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 12 / 18
17. Completion Time
Combination of strategies used: 9C-C-R
0
20000
40000
60000
80000
100000
120000
140000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
BOINC & SMALL BoT
0
5000
10000
15000
20000
25000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
BOINC & BIG BoT
0
10000
20000
30000
40000
50000
60000
70000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
BOINC & RANDOM BoT
0
5000
10000
15000
20000
25000
30000
35000
40000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
XWHEP & SMALL BoT
0
1000
2000
3000
4000
5000
6000
7000
8000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10
SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
XWHEP & BIG BoT
1000
2000
3000
4000
5000
6000
7000
8000
SETI
N
D
G
5K
LY
OG
5K
G
RESPO
T10
SPO
T100
Completiontime(s)
BE-DCI
No SpeQuloS
SpeQuloS
XWHEP & RANDOM BoT
→ Up to 9x speedup
→ Depend on middleware used, BE-DCI volatility
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 13 / 18
18. Completion Time Prediction
→ User can ask prediction at any moment of BoT execution
Predicted completion time:
tp = α ×
t(r)
r
Current completion ratio: r
Time elapsed since submission: t(r)
α: adjustment factor, depend on execution environment:
DG server & middlware
Application & BoT size
→ Adjusted after BoT execution to minimize difference w/ completion time
observed
Statistical uncertainty (±x%): Success rate of prediction vs previous execution
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 14 / 18
19. Prediction Results
Completion Time Predication:
Made at 50% of BoT execution
Uncertainty: ± 20%
α adjusted after 30 execution w/ same BD-DCI, middleware, BoT workload
BoT category & Middleware
SMALL BIG RANDOM
BE-DCI BOINC XWHEP BOINC XWHEP BOINC XWHEP Mixed
seti 100 100 100 82.8 100 87.0 94.1
nd 100 100 100 100 100 96.0 99.4
g5klyo 88.0 89.3 96.0 87.5 75 75 85.6
g5kgre 96.3 88.5 100 92.9 83.3 34.8 83.3
spot10 100 100 100 100 100 100 100
spot100 100 100 100 100 76 3.6 78.3
Mixed 97.6 96.1 99.2 93.5 89.6 65.3 90.2
→ Successful prediction in 9 cases out of 10
→ Lower results with heterogeneous BoT
→ Needs a learning phase, with same BoT (at least same app.), executed on
same BE-DCI.
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 15 / 18
20. SpeQuloS Deployment in European Desktop Grid Initiative
EDGI project: Bringing European Desktop Grids computing resources to scientific
communities.
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 16 / 18
21. Conclusion
BE-DCIs: “Low-cost” solution but poor QoS (tail effect)
SpeQuloS: Use Cloud resources to improve QoS delivered to BE-DCI users
Efficiently removes the tail problem
→ Speed up BoT execution
→ Only require few % of workload to be executed on Cloud
Enable completion time prediction for users
→ A step towards BE-DCIs usability in computing landscape ?
Future work:
Better strategies to anticipate problems (tail effect)
Analysis from users feedback in SpeQuloS deployments
S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 17 / 18
22. S. Delamare, G. Fedak, D. Kondo and O. Lodygensky (LIP/CNRS, Univ. Lyon, France, LIP/INRIA, Univ. Lyon, France, LIG/INRIA, Univ. Grenoble, France, LAL/SpeQuloS HPDC’12 18 / 18