Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST...
-
Upload
douglas-parks -
Category
Documents
-
view
223 -
download
0
Transcript of Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST...
![Page 1: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/1.jpg)
Wei Baiwith Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang
SING Group @ HKUST
Information-Agnostic Flow Scheduling for Commodity Data Centers
1SJTU, June 1st, 2015
![Page 2: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/2.jpg)
2
Data Centers
![Page 3: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/3.jpg)
3
This talk is about the transport inside the data center
![Page 4: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/4.jpg)
4
Data Center Networks
• High bandwidth• Low latency• Commodity switches• Single administrative domain
![Page 5: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/5.jpg)
5
Data Center Transport
• Cloud applications– Desire low latency for short messages
• Goal: Minimize flow completion time (FCT)– Many flow scheduling proposals…
![Page 6: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/6.jpg)
6
The State-of-the-art Solutions
• PDQ [SIGCOMM’12]
• pFabric [SIGCOMM’13]
• PASE [SIGCOMM’14]• …
All assume prior knowledge of flow size information to approximate ideal preemptive Shortest Job First (SJF) with customized network elements
![Page 7: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/7.jpg)
7
The State-of-the-art Solutions
• PDQ [SIGCOMM’12]
• pFabric [SIGCOMM’13]
• PASE [SIGCOMM’14]• …
All assume prior knowledge of flow size information to approximate ideal preemptive Shortest Job First (SJF) with customized network elements
Not feasible for some applications
![Page 8: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/8.jpg)
8
The State-of-the-art Solutions
• PDQ [SIGCOMM’12]
• pFabric [SIGCOMM’13]
• PASE [SIGCOMM’14]• …
All assume prior knowledge of flow size information to approximate ideal preemptive Shortest Job First (SJF) with customized network elements
Hard to deploy in practice
![Page 9: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/9.jpg)
9
Question
Without prior knowledge of flow size information, how to minimize FCT in commodity data centers?
![Page 10: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/10.jpg)
10
Design Goal 1
Without prior knowledge of flow size information, how to minimize FCT in commodity data centers?
Information-agnostic: not assume a priori knowledge of flow size information available from the applications
![Page 11: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/11.jpg)
11
Design Goal 2
Without prior knowledge of flow size information, how to minimize FCT in commodity data centers?
FCT minimization: minimize average and tail FCTs of short flows & not adversely affect FCTs of large flows
![Page 12: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/12.jpg)
12
Design Goal 3
Without prior knowledge of flow size information, how to minimize FCT in commodity data centers?
Readily-deployable: work with existing commodity switches & be compatible with legacy network stacks
![Page 13: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/13.jpg)
13
Question
Without prior knowledge of flow size information, how to minimize FCT in commodity data centers?
Our answer: PIAS
![Page 14: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/14.jpg)
14
PIAS’S DESIGN
![Page 15: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/15.jpg)
15
Design Rationale
• PIAS performs Multi-Level Feedback Queue (MLFQ) to emulate Shortest Job First
Priority 1
Priority 2
Priority K
……
High
Low
![Page 16: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/16.jpg)
16
Design Rationale
• PIAS performs Multi-Level Feedback Queue (MLFQ) to emulate Shortest Job First
Priority 1
Priority 2
Priority K
……
![Page 17: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/17.jpg)
17
Simple Example Illustrating PIAS
Congestion
![Page 18: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/18.jpg)
18
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 1 with 10 packets and flow 2 with 2 packets arrive
Priority 3
![Page 19: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/19.jpg)
19
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 1 and 2 transmit simultaneously
Priority 3
![Page 20: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/20.jpg)
20
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 2 finishes while flow 1 is demoted to priority 2
Priority 3
![Page 21: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/21.jpg)
21
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 3 with 2 packets arrives
Priority 3
![Page 22: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/22.jpg)
22
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 3 and 1 transmit simultaneously
Priority 3
![Page 23: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/23.jpg)
23
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 3 finishes while flow 1 is demoted to priority 3
Priority 3
![Page 24: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/24.jpg)
24
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 4 with 2 packets arrives
Priority 3
![Page 25: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/25.jpg)
25
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Flow 4 and 1 transmit simultaneously
Priority 3
![Page 26: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/26.jpg)
26
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Priority 3
Flow 4 finishes while flow 1 is demoted to priority 4
![Page 27: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/27.jpg)
27
Simple Example Illustrating PIAS
Priority 1
Priority 2
Priority 4
Priority 3
Eventually, flow 1 finishes in priority 4
With MLFQ, PIAS can emulate Shortest Job First without prior knowledge of flow size information
![Page 28: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/28.jpg)
28
How to implement?
Priority 1
Priority 2
Priority K
……
• Strict priority queueing on switches• Packet tagging as a shim layer at end hosts
• priorities:
• demotion thresholds: • The threshold to demote priority
from to is
![Page 29: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/29.jpg)
29
How to implement?
Priority 1
Priority 2
Priority K
……
• Strict priority queueing on switches• Packet tagging as a shim layer at end hosts
1
![Page 30: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/30.jpg)
30
How to implement?
Priority 1
Priority 2
Priority K
……
• Strict priority queueing on switches• Packet tagging as a shim layer at end hosts
![Page 31: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/31.jpg)
31
How to implement?
Priority 1
Priority 2
Priority K
……
• Strict priority queueing on switches• Packet tagging as a shim layer at end hosts
2
![Page 32: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/32.jpg)
32
• Thresholds depend on:– Flow size distribution– Traffic load
• Solution:– Solve a FCT minimization problem to calculate
demotion thresholds • Problem:– Traffic is highly dynamic
Determine Thresholds
Traffic variations -> Mismatched thresholds
![Page 33: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/33.jpg)
33
Impact of Mismatches
High
Low
10MB
10MB
20KB
![Page 34: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/34.jpg)
34
• When the threshold is perfect (20KB)
Impact of Mismatches
![Page 35: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/35.jpg)
35
• When the threshold is too small (10KB)
Impact of Mismatches
Increased latency for short flows
![Page 36: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/36.jpg)
36
• When the threshold is too large (1MB)
Impact of Mismatches
Increased latency for short flows
![Page 37: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/37.jpg)
37
• When the threshold is too large (1MB)
Impact of Mismatches
Increased latency for short flows
Leverage ECN to keep low buffer occupation
![Page 38: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/38.jpg)
38
• When the threshold is too small (10KB)
Handle Mismatches
If we enable ECN
![Page 39: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/39.jpg)
39
• When the threshold is too small (10KB)
Handle Mismatches
ECN can keep low latency
![Page 40: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/40.jpg)
40
• When the threshold is too large (1MB)
Handle Mismatches
If we enable ECN
![Page 41: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/41.jpg)
41
• When the threshold is too large (1MB)
Handle Mismatches
ECN can keep low latency
![Page 42: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/42.jpg)
42
PIAS in 1 Slide
• PIAS packet tagging– Maintain flow states and mark packets with priority
• PIAS switches– Enable strict priority queueing and ECN
• PIAS rate control– Employ Data Center TCP to react to ECN
![Page 43: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/43.jpg)
43
Testbed Experiments
• PIAS prototype– http://sing.cse.ust.hk/projects/PIAS
• Testbed Setup– A Gigabit Pronto-3295 switch– 16 Dell servers
• Benchmarks– Web search (DCTCP paper)– Data mining (VL2 paper)– Memcached
![Page 44: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/44.jpg)
44
Small Flows (<100KB)
Web Search Data Mining
Compared to DCTCP, PIAS reduces average FCT of small flows by up to 47% and 45%
47%45%
![Page 45: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/45.jpg)
45
NS2 Simulation Setup
• Topology– 144-host leaf-spine fabric with 10G/40G links
• Workloads– Web search (DCTCP paper) – Data mining (VL2 paper)
• Schemes– Information-agnostic: PIAS, DCTCP and L2DCT – Information-aware: pFabric
![Page 46: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/46.jpg)
46
Overall Performance
Web Search Data Mining
PIAS has an obvious advantage over DCTCP and L2DCT in both workloads.
![Page 47: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/47.jpg)
47
Small Flows (<100KB)
Simulations confirm testbed experiment results
Web Search Data Mining
40% - 50% improvement
![Page 48: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/48.jpg)
48
Comparison with pFabric
PIAS only has 4.9% performance gap to pFabric for small flows in data mining workload
Web Search Data Mining
![Page 49: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/49.jpg)
49
Conclusion
• PIAS: practical and effective– Not assume flow information from applications
– Enforce Multi-Level Feedback Queue scheduling
– Use commodity switches & legacy network stacks
Information-agnostic
FCT minimization
Readily deployable
![Page 50: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/50.jpg)
50
Thanks!
![Page 51: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/51.jpg)
51
Starvation
• Measurement– 5000 flows, 5.7 million MTU-sized packets– 200 timeouts, 31 two consecutive timeouts
• Solutions– Per-port ECN pushes back high priority flows when
many low priority flow get starved– Treating a long-term starved flow as a new flow
![Page 52: Wei Bai with Li Chen, Kai Chen, Dongsu Han, Chen Tian, Hao Wang SING Group @ HKUST Information-Agnostic Flow Scheduling for Commodity Data Centers 1 SJTU,](https://reader036.fdocuments.net/reader036/viewer/2022062314/56649f185503460f94c2ee0b/html5/thumbnails/52.jpg)
52
Persistent Connections
• Solution: periodically reset flow states based on more behaviors of traffic– When a flow idles for some time, we reset the
bytes sent of this flow to 0.– Define a flow as packets demarcated by incoming
packets with payload within a single connection