MetaCartSign in to MyCiteSeer

Include Citations | Advanced Search | Help

Include Citations | Advanced Search | Help

  a and Per Gunningberg

Download:
pdf | ps
by Kjersti Moldeklev
ftp://ftp.lirmm.fr/pub/odbi/moldeklev1.ps.gz
Add To MetaCart

Abstract:

The implementation of protocols, such as TCP/IP, and their integration into the operating system environment is very decisive for protocol performance. Putting TCP on high-speed networks, e.g. ATM, with large maximum transmission units causes the TCP maximum segment size to be relatively large. What Nagle's algorithm consider a "small " segment is not small anymore which affects the TCP throughput. We report on TCP/IP throughput performance measurements for various sizes of send and receive socket buffers, using two Sun IPX machines running SunOS 4.1.1 connected to FORE System's ATM network. For some common combinations of socket buffer sizes we observe a dramatic performance drop to less than 3 % of normal throughput. The drop is caused by a deadlock situation in the TCP connection which is resolved by the 200 ms spaced timer generated TCP acknowledgment. The factors which in combination force the TCP connection into deadlocks are a large maximum segment size, asymmetry of the socket buffer sizes, use of Nagle's algorithm, the delayed acknowledgment strategy, the sequence of actions on acknowledgment reception, and at last the socket layer optimization for efficient memory management. We explain what causes the deadlock situations, discuss some alternatives to avoid or prevent the situations and present measurement results from proposed changes in the implementation.

Citations

1790 Congestion avoidance and control – Jacobson - 1988
409 TCP/IP Illustrated, Volume 1: The Protocols – Stevens - 1994
273 Requirements for Internet Hosts – Communication Layers – Braden - 1989
208 C.—“Gigabit Networking – Partridge - 1994
206 The packet filter: An efficient mechanism for user-level network code – Mogul, Rashid, et al. - 1987
36 Internet working with TCP/IP, Principles, Protocols, and Architectures. Upper Saddle River – COMER - 2000
36 TCP Extensions for High Performance", RFC 1323 – Jacobson, Braden, et al. - 1992
35 Modularity and efficiency in protocol implementation – Clark - 1982
28 Window and acknowledgment strategy – Clark - 1982
27 Is Layering Harmful – Crowcroft, Wakeman, et al. - 1992
22 The TCP Maximum Segment Size and Related Topics,” RFC 879 – Postel - 1983
16 Congestion Control in TCP/IP Internetworks – Nagle - 1984
16 Requirements for Internet Hosts – Communication Layers," Internet Request for Comments 1122 – Braden - 1989
11 Congestion Avoidance and Control – Jacobsen - 1988
3 D.Borman "TCP extensions for high-performance – Jacobsen, Braden - 1992
2 Transmission Control Protocol, protocol specification – Postel - 1981
2 et al. 4.3 BSD Unix operating system – Leffler - 1989
2 et al.: "Is layering harmful – Crowcroft - 1992
1 Internet protocol, protocol specification – Postel - 1981
1 Deadlock situations in TCP over ATM", To appear – Moldeklev, Gunningberg - 1994