Why do UDP latencies spike?

Submitted by dstockdill on Thu, 01/17/2013 - 12:02

As described in TCP/IP Illustrated, Volume 1, by W. Richard Stevens, Section 11.9, ARP entries that expire may introduce high latencies on the next send. Entry expiration details are system-dependent; for instance, a packet sent to a particular address may refresh the timeout.

On a VxWorks target, the RTI latency test exhibited this behavior as an unusually high maximum latency for one trial (see the third row from the bottom):

bytes, stdev us,ave us, min us, 50% us, 90% us, 99% us, 99.99%,  max us
------,-------, -------,------, ------, ------, ------, ------, -------
    16,   2.6,   192.9,  175.6,  193.0,  194.0,  195.0,  242.0,   289.8
    32,   2.2,   197.8,  180.1,  197.0,  198.0,  199.0,  246.0,   308.9
    64,   2.4,   208.3,  192.8,  208.0,  208.0,  212.0,  258.0,   307.5
   128,   2.9,   228.8,  217.5,  228.0,  228.0,  236.0,  278.0,   342.0
   256,   3.1,   274.3,  264.0,  272.0,  274.0,  290.0,  322.0,   385.2
   512,   3.0,   363.7,  351.9,  362.0,  364.0,  376.0,  412.0,   468.5
  1024,  94.5,   542.2,  535.8,  540.0,  540.0,  556.0,  592.0, 67350.9
  2048,   3.3,   795.8,  790.8,  792.0,  792.0,  804.0,  844.0,   923.4
  4096,   3.6,  1169.8, 1162.8, 1160.0, 1170.0, 1180.0, 1240.0,  1395.1 

One workaround is to increase the ARP timeout. The workaround is again system-dependent; for example, on VxWorks 5.5, the timeout is defined by the global variable, arpt_keep, which defaults to 1200 (20 minutes).

Product:
Platform:
Keywords: