The Implications of Universal Redundant Traffic Elimination Ashok Anand Archit Gupta Aditya Akella University of Wisconsin Madison Srinivasan Seshan Carnegie Mellon University ID: 650773
Download Presentation The PPT/PDF document "Packet Caches on Routers:" 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
Packet Caches on Routers: The Implications of Universal Redundant Traffic Elimination
Ashok Anand, Archit Gupta, Aditya AkellaUniversity of Wisconsin, MadisonSrinivasan SeshanCarnegie Mellon University Scott Shenker University of California, Berkeley
1Slide2
Redundant Traffic in the Internet
Lots of redundant traffic in the InternetRedundancy due to…Identical objectsPartial content match (e.g. page banners)Application-headers…2
Same
c
ontent
t
raversing
same set of links
Time
T
Time
T + 5Slide3
Redundancy Elimination
Object-level cachingApplication layer approaches like Web proxy caches Store static objects in local cache[Summary Cache: SIGCOMM 98, Co-operative Caching: SOSP 99]Packet-level caching [Spring et. al: SIGCOMM 00]WAN Optimization Products: Riverbed, Peribit, Packeteer, ..3
Packet-Cache
Packet-Cache
Access link
Internet
Enterprise
Packet-level caching is better than object-level cachingSlide4
Benefits of Redundancy EliminationReduces bandwidth usage cost
Reduces network congestion at access links Higher throughputsReduces in transfer completion times4Slide5
Towards Universal RE
However, existing RE approaches apply only to point deploymentsE.g. at stub network access links, or between branch officesThey only benefit the system to which they are directly connected.Why not make RE a native network service that everyone can use?5Slide6
Our Contribution
Universal redundancy elimination on routers is beneficialRe-designing the routing protocol to be redundancy aware gives furthermore benefitsPractical to implement redundancy elimination6Slide7
Internet2
Universal Redundancy Elimination At All Routers
Total packets with universal RE= 12
(ignoring tiny packets)
Upstream router removes redundant bytes
.
Downstream router reconstructs full packet
7
Total packets
w/o RE = 18
Wisconsin
Berkeley
CMU
33%
Packet cache
at every routerSlide8
Benefits of Universal Redundancy Elimination
Subsumes benefits of point deploymentsAlso benefits Internet Service ProvidersReduces total traffic carried better traffic engineeringBetter responsiveness to sudden overload (e.g. flash crowds)Re-design network protocols with redundancy elimination in mind Further enhance the benefits of universal RE8Slide9
Redundancy-Aware Routing
Total packets with RE + routing= 10
(Further 20% benefit )
9
Total packets with RE = 12
Wisconsin
Berkeley
CMU
45%
ISP needs information of traffic similarity between CMU and Berkeley
ISP needs to compute redundancy-aware routesSlide10
Redundancy-Aware RoutingIntra-domain Routing for ISP
Every N minutesEach border router computes a redundancy profile for the first Ts of the N-minute intervalEstimates how traffic is replicated across other border routersHigh speed algorithm for computing profilesCentrally compute redundancy-aware routesRoute traffic for next N minutes on redundancy-aware routes. Redundancy elimination is applied hop-by-hop 10Slide11
CMU
Redundancy Profile Example11Internet2
Data
unique,pitsburgh
= 30 KB
Data
unique,Berkeley
= 30 KB
Data
shared
= 20 KB
11
Wisconsin
Berkeley
Total
CMU
= 50 KB
Total
Berkeley
= 50 KBSlide12
Centralized Route Computation
Linear ProgramObjective: minimize the total traffic footprint on ISP linksTraffic footprint on each link as latency of link times total unique content carried by the linkCompute narrow, deep trees which aggregate redundant traffic as much as possibleImpose flow conservation and capacity constraints12
Centralized
Platform
Route
c
omputationSlide13
Inter-domain RoutingISP selects neighbor AS and the border router for each destination
Goal: minimize impact of inter-domain traffic on intra-domain links and peering links.Challenges:Need to consider AS relationships, peering locations, route announcementsCompute redundancy profiles across destination ASesDetails in paper13Slide14
Trace-Based Evaluation
Trace-based studyRE + Routing: Redundancy aware routingRE: Shortest path routing with redundancy eliminationBaseline: Compared against shortest path routing without redundancy elimination Packet tracesCollected at University of Wisconsin access linkSeparately captured the outgoing traffic from separate group of high volume Web servers in University of WisconsinRepresents moderate-sized data centerRocketfuel ISP topologiesResults for intra-domain routing on Web server trace14Slide15
Benefits in Total Network Footprint
Average redundancy of this Web server trace is 50% using 2GB cacheATT topology2GB cache per routerCDF of reduction in network footprint across border routers of ATTRE gives reduction of 10-35%(RE + Routing) gives reduction of 20-45% 15Slide16
When is RE + Routing Beneficial?Topology effect
E.g., multiple multi-hop paths between pairs of border routersRedundancy profileLot of duplication across border routers16Slide17
Synthetic Trace Based Study
Synthetic trace for covering wide-range of situationsDuplicates striped across border routers in ISP (inter-flow redundancy)Low striping across border routers , but high redundancy with in traffic to a border router (intra-flow-redundancy)Understand topology effect17Slide18
Benefits in Total Network Footprint
Synthetic trace, average redundancy = 50%ATT (7018) topologyTrace is assumed to enter at Seattle RE + Routing, is close to RE at high intra-flow redundancy, 50% benefitRE has benefit of 8% at zero intra-flow redundancyRE + Routing, gets benefit of 26% at zero intra-flow redundancy.18Slide19
Benefits in Max Link Utilization
Link capacities either 2.5 or 10 GbpsComparison against traditional OSPF based traffic engineering (SP-MaxLoad)RE offers 1-25% lower maximum link load . RE + Routing offers 10-37% lower maximum link load. Max link Utilization = 80%, for SP-MaxLoad19Slide20
Evaluation SummaryRE significantly reduces network footprint
RE significantly improves traffic engineering objectivesRE + Routing further enhances these benefitsHighly beneficial for flash crowd situationsHighly beneficial in inter-domain traffic engineering20Slide21
Implementing RE on Routers
21Fingerprint tablePacket store
Fingerprint s
Main operations
Fingerprint computation
Easy, can be done with CRC
Memory operations, Read and WriteSlide22
High Speed Implementation
Reduced the number of memory operations per packetFixed number of fingerprints (<10 per packet)Used lazy invalidation of fingerprint for packet evictionOther optimizations in paperClick-based software prototype runs at 2.3 Gbps (approx. OC 48 speed ).22Slide23
Summary
RE at every router is beneficial ( 10-50%)Further benefits (10-25%) from redesigning routing protocol to be redundancy-aware.OC48 speed attainable in software23Slide24
Thank you
24Slide25
Backup
25Slide26
Flash Crowd Simulation
Flash Crowd: Volume increases at one of the border routersRedundancy ( 20% -> 50%)Inter Redundancy Fraction (0.5 -> 0.75)Max Link Utilization without RE is 50%Traditional OSPF traffic engineering gets links at 95% utilization at volume increment factor > 3.5Whereas SP-RE at 85% , and RA further lower at 75%26Slide27
Impact of Stale Redundancy Profile
RA relies on redundancy profiles.How stable are these redundancy profiles ?Used same profile to compute the reduction in network footprint at later times ( with in an hour)RA-stale is quite close to RA27Slide28
High Speed Implementation
Use specialized hardware for fingerprint computationReduced the number of memory operations per packetNumber of memory operations is function of number of fingerprints. Fixed the number of sampled fingerprintsDuring evicting packet, explicit invalidating fingerprint require memory operations. Used lazy invalidationFingerprint pointer is checked for validation as well as existence.Store packet-table and fingerprint-table in DRAM for high-speedUsed Cuckoo Hash-table. As simple-hash based fingerprint table is too large to fit in DRAM28Slide29
Base Implementation Details (Spring et. al)Compute fingerprints per packet and sample them
Insert packet into packet storeCheck for existence of fingerprint pointer to any packet, for match detection. Encode the match region in the packet.Insert each fingerprint into Fingerprint table.As store becomes full, evict the packet in FIFO mannerAs a packet gets evicted, invalidate its corresponding fingerprint pointers29