A Reconfigurable Wireless Datacenter Fabric using FreeSpace Optics Navid Hamedazimi Zafar Qazi Himanshu Gupta Vyas Sekar Samir Das Jon Longtin Himanshu Shah Ashish ID: 277937
Download Presentation The PPT/PDF document "FireFly:" 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
FireFly: A Reconfigurable Wireless Datacenter Fabric using Free-Space Optics
Navid Hamedazimi, Zafar Qazi, Himanshu Gupta, Vyas Sekar, Samir Das, Jon Longtin, Himanshu Shah, Ashish Tanwer
ACM SIGCOMM 2014Slide2
Datacenter network design is hard!
CostPerformanceCabling
Expandability
Energy
Cooling
Adaptability
2Slide3
Existing Data Center Network Architectures
…
Over subscribed
(e.g. simple tree)
Augmented (e.g.
cThrough
)
…
u
Over provisioned
(e.g.
FatTree
, Jellyfish)
…
3Slide4
Our Vision : FireFly4
Coreless
Wireless
Steerable
ToR
switch
FireFly
Controller
Steerable
LinksSlide5
Potential Benefits of This VisionCost
PerformanceCabling
Expandability
Energy
Cooling
Adaptability
5
Wireless
Coreless
SteerableSlide6
Challenges in Realizing the Vision6
FireFlyController
ToR
switch
Steerable
FSOs
Steerable wireless links
Network Design
Network Management
FireFly
shows this vision is feasibleSlide7
OutlineMotivationSteerable Wireless Links
Network DesignNetwork ManagementEvaluation7Slide8
Why FSO instead of RF?8
RF (e.g. 60GHZ)
FSO (Free Space optical)
Wide beam
High interference
Limited active links
Limited Throughput
Narrow beam
Zero interference
No limit on active links
High ThroughputSlide9
9Today’s FSO
Cost: $15K per FSOSize: 3 ft³Power: 30wNon steerable Current: bulky, power-hungry, and expensive
Required: small, low power and low expenseSlide10
Why Size, Cost, Power Can be Reduced? 10
Traditional use : outdoor, long haul High power Weatherproof Data centers: indoor, short haul Feasible roadmap via commodity fiber optics E.g. Small form transceivers (Optical SFP)Slide11
FSO Design Overview
11SFPfiber optic cables
Diverging beam
Lens focal distance
large cores (> 125 microns) are more robust
Large core fiber optic cables
Parallel beam
lens
Focusing lens
Collimating lensSlide12
12Steerability
CostSizePowerNot Steerable
FSO design using SFP
V
ia Switchable mirrors
or
Galvo
mirrors
Shortcomings of current FSOsSlide13
Steerability via Switchable Mirror13
A
Ceiling mirror
B
C
Switchable Mirror: glass
mirror
Electronic control, low latency
SM in “mirror”
modeSlide14
Steerability via Galvo Mirror
14
A
Ceiling mirror
B
C
Galvo
Mirror: small rotating mirror
Very low latency
Galvo
MirrorSlide15
FSO Prototype in Data center15
Fiber holder and lens
MirrorSlide16
FSO Link Performance
6
mm
6
mm
16
FSO link is as robust as a wired link
Effect of vibrations, etc.
6mm movement tolerance
Range up to 24m testedSlide17
OutlineMotivationSteerable Wireless Links
Network DesignNetwork ManagementEvaluation17Slide18
How to design FireFly network?
18Goals: Robustness to current and future trafficBudget & Physical Constraints
Design parametersNumber of FSOs?Number of steering mirrors?Initial mirrors’ configurationPerformance metricDynamic bisection bandwidthSlide19
FireFly Network Design
19# of FSOs = # of Servers# of Switchable Mirrors = [10-15] for up to 512 racks
or# of Galvo Mirrors = 1 per FSOMirror Configuration = Random graph
less than ½ the ports of FatTree
Projected Cost: 40% to 60% lower than
FatTreeSlide20
OutlineMotivationSteerable Wireless Links Network Design
Network ManagementEvaluation20Slide21
Network Management Challenges
21
Reconfiguration
Traffic engineering
Topology
control
Correctness during flux
ToR
switch
FireFly
Controller
Steerable
FSOs
Ceiling MirrorSlide22
FireFly Reconfiguration Algorithm22
Joint optimization problem
Decouple
Traffic
engineering
Topology control
Above is done periodically
In addition: Trigger-based reconfiguration
E.g. Create direct link for large flows
Massive ILP
Max-flow
, greedy
Weighted MatchingSlide23
Correctness Problems During Flux23
Connectivity
Black Holes
Latency
A
B
A
B
A
B
C
C
CSlide24
Simple Rules To Ensure Correctness24
Disallow
deactivations that
disconnect the network.
Stop using a link before deactivating it
Start using a link only after activating it
“
Small”
gap
between
reconfigurationsSlide25
OutlineMotivationSteerable Wireless Links Network Design
Network ManagementEvaluation25Slide26
FireFly EvaluationPacket-level
Flow-level (for large scale networks)Evaluation of network in-fluxEvaluation of Our Heuristics 26Slide27
27
FireFly is comparable to FatTree with less than ½ the portsFlow completion time better than FatTree
FireFly
ThroughputSlide28
ConclusionsVision: Extreme DC network architectureFully Steerable, No core switches, All-wireless inter-rackUnprecedented benefits:
No Cabling, Adapt to traffic patterns, Less clutter Firefly shows a viable proof pointPractical steerable FSO for datacentersPractical network design and management heuristicsClose to fat tree performance over several workloadsLess than half of FatTree portsJust a start .. Many directions for improvement
28