Congestion Control, Traffic Management and QoS (Lecture 5)

Scribed by Daniel Lau Jitendra Padhye et. al.: Modeling TCP Throughput: A Simple Model and its Empirical Validation Summary The purpose of the paper is to provide an accurate model of TCP by analyzing fast retransmit and timeouts. Creating models for TCP is especially important in order to define TCP-friendly characteristics for non-TCP flows. Several [...]

Data Center (lecture 6)

Scribed by Arti Gupta: Albert Greenberg et. al, ,VL2: A Scalable and Flexible Data Center Network, SIGCOMM 2009 Summary: This paper by the Microsoft team briefly describes the current problems faced by the data centers like : Limited server-server capacity : As we move higher up in the hierarchy of switches and routers, we see [...]

Data Center (Lecture 6)

Albert Greenberg et al., VL2: A Scalable and Flexible Data Center Network, SIGCOMM 2009 Scribed by James Hongyi Zeng (hyzeng_at_stanford.edu) Summary: In this paper, Albert Greenberg et al describe VL2, a new data center network architecture. The design goal of VL2 is 1) Uniform high capacity, 2) Performance isolation, and 3) Layer-2 semantics. Before diving [...]

Congestion Control, Traffic Management and QoS (Lecture 5)

Jitendra Padhye et. al., Modeling TCP Throughput: A Simple Model and its Empirical Validation, ACM SIGCOMM 1998 Scribed by James Hongyi Zeng (hyzeng_at_stanford.edu) Summary: In this paper, Jitendra Padhye et.al. model the TCP throughput analytically as a function of  packet loss rate, round trip time (RTT), and maximum window size. They empirically validate the result [...]

Congestion Control, Traffic Management and QoS (Lecture 4)

Scribed by: kavindaw Congestion Avoidance and Control Summary When the Internet suffered congestion collapses in October of 1986, it was clear that there was something wrong with TCP on existing hosts.  Jacobson and Karels clearly pointed out that it was not the protocol itself, but how it was implemented that caused massive problems.  They argued [...]

Congestion control, traffic management and QoS (Lecture 4)

Scribed by Daniel Sanchez Congestion Avoidance and Control, V. Jacobson and M.J. Karels Summary This paper explains a set of improvement to the congestion control algorithms of 4.3BSD TCP. The ideas are based in the principle of maintaining a conservative flow: for a connection in equilibrium, a new packet is not injected into the network [...]

Data Center (Lecture 6)

Scribed by: Harrison Ting Albert Greenberg et al., VL2: A Scalable and Flexible Data Center Network, SIGCOMM 2009 Summary: Data centers, especially for cloud services, have many services for which it is desirable to share servers.  Typical data centers have tree-like network configurations with highly oversubscribed nodes, thus limiting communication between servers. The authors from [...]

Naming and Routing (Lecture 4)

General Discussion Most widely used TCP flavor: NewReno with Selective ACK TCP flavors vary with OS implementation Link Utilization – No buffer Maximum Utilization of link < 100% When 2 packets are sent, no buffer, one packet dropped – no buffer 75% average utilization with single flow Buffer size = 2 times bandwidth-delay product → [...]

Naming and Routing (Lecture 4)

Scribed by Frank Nothaft Congestion Avoidance and Control Summary: This paper discusses the redevelopment of TCP motivated by the occurrence of a “congestion collapse” where a high level of congestion caused throughput over an example line to drop from 32 Kbps to 40 bps. Specifically, this paper discusses the logic behind the implementation of slow-start [...]

Data Centers (Lecture 6)

Scribed by Izaak Rubin (izaak@stanford.edu) Albert Greenberg et. al., VL2: A Scalable and Flexible Data Center Network, SIGCOMM 2009 Summary: This paper proposes a way to create the illusion of a large virtual layer 2 (VL2) network, as if all nodes were connected by one big switch.  Such a network is useful in large data [...]