Commit cf8b4e52 authored by Pramod R Sanaga's avatar Pramod R Sanaga

Literature survey of papers that use network paths in their eval section.

parent fd7f5e19
Survey of papers from top conferences(SIGCOMM,OSDI & NSDI,2004-2007) which use
some kind of network topology with or without background traffic in their
evaluation.
Original Source - Amin's Usenix '08 paper - "Does Background Traffic Matter?"
Out of a total of 23 papers:
1) 4-5 papers used the correct setup - with separation of capacity vs abw,
background traffic and appropriate queue sizes.
2) Around 8 papers used the wrong kind of setup ( in one way or the other,
Eg: no background traffic, no mention of queue sizes etc ).
3) The other papers are not relevant to this topic.
1) SIGCOMM '06: Minimizing Churn in distributed systems
Simulations with a discrete event simulator - mainly for demonstrating churn in a DHT. No background traffic. Latencies + capacities
are mentioned for links ( latencies are taken from publicly available traces ).
It appears that each link was modeled on its own without bottlenecks & without reactivity. ( and no mention of the queue sizes ).
2) SIGCOMM '06: Speakup ( DDoS defense by offense )
Used Emulab - Òvarious emulated topologiesÓ
Most important: Dumbbell shaped link(40Mbps) with a server on one side and 30 clients on the other with access links of 2 Mbps each.
No background traffic was mentioned. The paths are probably modeled as single links.
3) SIGCOMM '04: Modeling and Performance Analysis of BitTorrent-Like Peer-to-Peer Networks
Simulated a P2P network with a fixed upload/download bandwidth per client. No mention of background traffic and capacity/abw separation or queue sizes.
4) SIGCOMM '04: Mercury: Supporting Scalable Multi-Attribute Range Queries
Simulation of large topologies ( 50k nodes ) for the purpose of routing. Does not have queuing delays or links. Our method is not applicable here because
sending traffic is not important in this expt.
5) OSDI 04: FUSE: Lightweight Guaranteed Distributed Failure Notification.
Used Modelnet with OC3 ( 155 Mbps ) and T3 links and realistic latencies. The topologies are real with 100k links and 100k nodes. No mention of background traffic or the queue sizes on the links.
6) NSDI 05: Quorum: Flexible Quality of Service for Internet Services
No mention of any topology: At a high level, Internet clients are connect to a load balancer, which leads to a cluster of machines in a LAN.
ÒReplayed the Internet trace from some referred deploymentÓ. Probably used links to represent paths, without background traffic.
7) NSDI 05: Bandwidth-efficient Management of DHT Routing Tables
Simuation of large topologies ( 3k nodes ) for the purpose of routing. Does not have queuing delays or links. Our method is not applicable here because
sending traffic is not important in this expt.
8) NSDI 04: Macedon - not applicable, because they were comparing their implementation against the results from the original papers.
9) NSDI 04: TimeLine: A High Performance Archive for a Distributed Object Store
Not applicable - being compared to an earlier version of Thor. Network topology/abw is hardly a concern in this paper.
10) SIGCOMM '07: Structured streams
Simulates access links (1.5 Mbps). No background traffic - but probably ok because the bottleneck will probably be the access link.
11) NSDI 07: Exploiting Similarity for Multi-Source Downloads Using File Handprints
Uses Emulab to create topologies: Paths modeled with links and no capacity/abw separation. No mention of queue sizes and no background traffic.
12) SIGCOMM '04: Capprobe
Capacity estimation technique: Is interested in the characteristics of a single link - modeled with capacity and varying amount of tcp/udp cross traffic. No queue size, but that is probably ok.
13) NSDI '06: An Architecture for Internet Data Transfer
In some cases of the eval. section - ÒwanÓ like characteristics were probably modeled (on Emulab) with links. No background traffic or queue sizes or capacity/abw separation.
14) NSDI '06: Colyseus
Models paths as links with fixed latencies ( in Emulab ) - does not at all consider bandwidth/capacity, even though data is transferred between nodes.
15) NSDI '05: Trickles
Dumbbell topology: with a path being modeled using a link. No mention of queue sizes, although they do have background traffic.
16) NSDI '05: Debunking Some Myths About Structured and Unstructured Overlays
Not applicable. Routing queries in overlays.
17) SIGCOMM 2005: One more bit is enough ( VCP )
Uses different amounts of bottleneck bandwidth, with cross-traffic. They talk about the impact of router queue size too, although it is not explicitly mentioned as a parameter in the expts, which it probably is - Good setup.
18) NSDI '06: PCP
Used RON and simulation. Simulation setup is good with background traffic and queue size set to BDP.
19) SIGCOMM '05: TFRC
Simulation: background traffic and queue size BDP - Good setup
20) SIGCOMM '07: Emulating AQM from End Hosts
Simulation: background traffic and queue size BDP - Good setup
21) NSDI '06: Connection Handoff Policies for TCP Offload Network Interfaces.
Not applicable - testing wire throughput.
22) NSDI '06: Connection conditioning
Not applicable - not network related.
23) NSDI '06: Nakika
Not applicable - meta paper and not network related.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment