with Snappy Xiaoqi Chen Shir Landau Feibish Yaron Koral Ori Rottenstreich and Jennifer Rexford SIGCOMM SelfDN Workshop August 24 th 2018 Budapest ID: 680618
Download Presentation The PPT/PDF document "Catching t he Microburst Culprits" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Catching the Microburst Culprits with Snappy
Xiaoqi Chen, Shir Landau Feibish, Yaron Koral, Ori Rottenstreich and Jennifer Rexford
SIGCOMM
SelfDN
Workshop
August 24th, 2018Budapest, Hungary
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
1Slide2
Microbursts: Short Lived Traffic Bursts
Normal traffic rates are much lower than queue throughput
Buildup is normally minimal
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
2Slide3
Microbursts: Short Lived Traffic Bursts Occasional short lived traffic spikes
Cause significant queue buildup
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
3Slide4
Queue Buildup in Data Centers
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/184Slide5
Queue Buildup in Carrier Networks
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/185Slide6
Microbursts are expensive…
Network admins want to:avoid packet lossuse cheap switches high link utilizationssupport bursty workloadsSIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/186Slide7
Who caused the microburst?
The General Queue Occupancy Problem: What’s the size of each flow in the queue?Snappy solves:If a packet belongs to a heavy
flow
When queue is
long
Key
Count
1
5
1
1
2
1
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
7Slide8
Queue Occupancy Problem
3
Count
Key
1
1
1
The problem is hard! Simultaneous add and delete.
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
8Slide9
Queue Occupancy Problem
Count
Key
1
1
1
2
The problem is hard! Simultaneous add and delete.
Update
both
for
arrivals
and
departures
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
9Slide10
Solution: snapshots
Snappy maintains snapshots for short periods of incoming traffic. We then combine snapshots to estimate entire queue’s content.Observation 1: when queue is long, low relative error Observation 2: we
care about
heavy flows, not everyone
?
S1
S2
S3
S4
…
1
5
~
Count
Key
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
10Slide11
Round-Robin between Snapshots
Read
Read
Write
Clean
Read
Read
Read
Observation
3:
limited #snapshots needed.
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
11Slide12
Precision vs. Snapshot Size
Catching heavy flows:Using 4~8 snapshots is sufficient.SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/1812Slide13
In Queue Flow Size Estimation
Flow-size estimate:Low absolute error(~50kb)SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/1813Slide14
Summary & Future workProblem
Can’t add/delete simultaneouslyRestricted computation in data planeMicroburst is shortOUR SolutionUse snapshot to avoid deletion, combine snapshotsUse sketchImmediate action in data plane
Future Work
Deployment
on
Backbone
Variations on the queue model (Priority, non-FIFO)
Variations on the flow statistics (heavy
flow
groups)
Weighted action
s
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
14Slide15
Backup Slides
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/1815Slide16
Evaluation – Window size
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks8/24/1816Slide17
Protocol Independent Switch Architecture
Queuing metadata becomes availableSnappy snapshotslive here
R
R
W
C
R
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
17Slide18
Queuing and processing
Parser
Ingress
Pipe
Traffic
Manager
Queuing
Egress
Pipe
Deparser
Queue
Depth
info
becomes
available
Snappy
resides
here
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
18Slide19
Implementing Snappy on PISA:Approximation Using CM Sketch
Count-Min Sketch [CM ‘05]C columns
B Counters
f
+1
+1
+1
Register
Arrays
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
19Slide20
Backup: Snapshot Data Structure Residing in the data plane
Stage 1Stage 2Stage 3
Stage
4
Snap
1Row 1
+1
Snap
1
Row
2
+1
Snap
2
Row
1
Read
Snap
2
Row
2
Read
Packet
SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks
8/24/18
20