Difference between revisions of "Cluster:LowLatency"

From Earlham CS Department
Jump to navigation Jump to search
 
(23 intermediate revisions by 5 users not shown)
Line 1: Line 1:
−
==Low Latency TCP/IP for Linux notes==
+
==Internal Links==
 +
* [[Cluster:LLK_Poster_Structure|Poster Structure]] (Moved to CVS/LaTeX)
 +
* [[llk-packet-diagram|Packet Diagram]]
 +
* [[Running Timer Code]]
 +
* [[llk QA]]
 +
 
 +
==The Plan (updated January 28, 2006)==
 +
* Low-level benchmarks are netpipe (TCP) and pacgen (UDP).  High level benchmark is MrBayes.  Define run parameters for each set.  Setup scripts for PBS/Maui to run defined sets.  Run on cairo only.
 +
 
 +
* Formal literature search and review.  Use/add to the keywords in llk/poster/abstract.tex.  Places to look:
 +
** Beowulf archives
 +
** ACM
 +
** IEEE
 +
** Citeseer
 +
** Google Scholar
 +
** tp_timer
 +
** NIST
 +
 
 +
* Test tcp_low_latency - does it work?  If so, what does it do _exactly_
 +
 
 +
* Show that latency is in the network stack, either by reference or by test (preferrably the former since it will take less time).
 +
 
 +
==Notes==
 +
* BibTeX references are in /cluster/project/llk/llk.bib
 +
 
 
*Comparisons & Instrumentation
 
*Comparisons & Instrumentation
  
Line 9: Line 33:
 
***http://user.informatik.uni-goettingen.de/~kperf/
 
***http://user.informatik.uni-goettingen.de/~kperf/
 
***http://user.informatik.uni-goettingen.de/~kperf/gaug-ifi-tb-2005-03.pdf
 
***http://user.informatik.uni-goettingen.de/~kperf/gaug-ifi-tb-2005-03.pdf
 +
***http://christian-dickmann.de/telematik/presentation.pdf (their slideshow presentation)
  
 
**Comparing while varying link delay times
 
**Comparing while varying link delay times
Line 38: Line 63:
 
***Maximum Segment Size(MSS) http://rfc.activedomain.org/0500-0999/rfc0879.html
 
***Maximum Segment Size(MSS) http://rfc.activedomain.org/0500-0999/rfc0879.html
 
***Selective Acknowledgement(Sack) ftp://ftp.isi.edu/in-notes/rfc2018.txt
 
***Selective Acknowledgement(Sack) ftp://ftp.isi.edu/in-notes/rfc2018.txt
 +
***Timestamping http://www.faqs.org/rfcs/rfc1323.html
 +
**opensolaris code
 +
***http://cvs.opensolaris.org/source/xref/usr/src/uts/common/inet/
  
 
*Low Latency Kernel Option
 
*Low Latency Kernel Option
 
**net.ipv4.tcp_low_latency sysctl
 
**net.ipv4.tcp_low_latency sysctl
 
***/proc/sys/net/ipv4/tcp_low_latency
 
***/proc/sys/net/ipv4/tcp_low_latency
 +
**From /usr/src/linux/Documentation/networking/ip-sysctl.txt
 +
        tcp_low_latency - BOOLEAN
 +
        If set, the TCP stack makes decisions that prefer lower
 +
        latency as opposed to higher throughput.  By default, this
 +
        option is not set meaning that higher throughput is preferred.
 +
        An example of an application where this default should be
 +
        changed would be a Beowulf compute cluster.
 +
        Default: 0
 +
 +
== Ideas ==
 +
 +
* Reduce error checking to improve performance
 +
* Reduce number of memory copies, maybe by not being fully TCP/IP-compliant?
 +
* Look into STP (scheduled transfer protocol)
 +
* Check out the TCP_NODELAY sockopt:
 +
<pre>
 +
if(setsockopt(sockfd, proto->p_proto, TCP_NODELAY, &one, sizeof(one)) < 0)
 +
{
 +
  printf("NetPIPE: setsockopt: TCP_NODELAY failed! errno=%d\n", errno);
 +
  exit(556);
 +
}
 +
</pre>
 +
http://www.openldap.org/lists/openldap-devel/199907/msg00079.html
 +
 +
==Links==
 +
*[http://cluster.earlham.edu/bugzilla/buglist.cgi?query_format=specific&order=relevance+desc&bug_status=__open__&product=llk Current LLK bug list]
 +
 +
*[http://www.faqs.org/docs/Linux-HOWTO/Kernel-HOWTO.html Linux Kernel Howto]
 +
*[http://en.tldp.org/LDP/tlk/tlk.html The Linux Kernel Online Book]
 +
*[http://plg.uwaterloo.ca/~itbowman/papers/CS746G-a1.html Conceptual Architecture of the Linux Kernel (paper)]
 +
*[http://www.linuxhq.com/ The Linux Headquarters]
 +
*[http://kernelnewbies.org/ Kernel Newbies]
 +
*[http://www.ee.unimelb.edu.au/staff/lha/Linux_network_stack_walkthrough.html Walkthrough of the 2.4 network stack]
 +
*[http://www.spine-group.org/papers/stack/stack2.6.html 2.6 stack (Italian)]
 +
*[https://svn.gnumonks.org/trunk/doc/packet-journey-2.6.xml 2.6 packet journey (hard to read)]

Latest revision as of 19:10, 28 January 2006

Internal Links

The Plan (updated January 28, 2006)

  • Low-level benchmarks are netpipe (TCP) and pacgen (UDP). High level benchmark is MrBayes. Define run parameters for each set. Setup scripts for PBS/Maui to run defined sets. Run on cairo only.
  • Formal literature search and review. Use/add to the keywords in llk/poster/abstract.tex. Places to look:
    • Beowulf archives
    • ACM
    • IEEE
    • Citeseer
    • Google Scholar
    • tp_timer
    • NIST
  • Test tcp_low_latency - does it work? If so, what does it do _exactly_
  • Show that latency is in the network stack, either by reference or by test (preferrably the former since it will take less time).

Notes

  • BibTeX references are in /cluster/project/llk/llk.bib
  • Comparisons & Instrumentation
  • Low Latency Kernel Option
    • net.ipv4.tcp_low_latency sysctl
      • /proc/sys/net/ipv4/tcp_low_latency
    • From /usr/src/linux/Documentation/networking/ip-sysctl.txt
       tcp_low_latency - BOOLEAN
       If set, the TCP stack makes decisions that prefer lower
       latency as opposed to higher throughput.  By default, this
       option is not set meaning that higher throughput is preferred.
       An example of an application where this default should be
       changed would be a Beowulf compute cluster.
       Default: 0

Ideas

  • Reduce error checking to improve performance
  • Reduce number of memory copies, maybe by not being fully TCP/IP-compliant?
  • Look into STP (scheduled transfer protocol)
  • Check out the TCP_NODELAY sockopt:
 if(setsockopt(sockfd, proto->p_proto, TCP_NODELAY, &one, sizeof(one)) < 0)
 {
   printf("NetPIPE: setsockopt: TCP_NODELAY failed! errno=%d\n", errno);
   exit(556);
 }

http://www.openldap.org/lists/openldap-devel/199907/msg00079.html

Links

Retrieved from "https://wiki.cs.earlham.edu/index.php?title=Cluster:LowLatency&oldid=5364"