> there still seems to be a slight bump when electric fence was enabled; > and looking at the results some of the extra failures do show the > crypto slowdown. > > To test this theory, I'll set up the next test run on > testing.libreswan.org so that EFENCE is disabled.
Confirmed. While the the last run at http://testing.libreswan.org/results/ isn't yet finished the results are already clear. Off hand, I can think of two reasons: - IKEv1 hammers malloc() et.al. more than IKEv2 While I have my doubts, I've no real evidence either way. Anyone worried enough to compare the number of malloc() et.al. calls used by IKEv1 vs IKEv2 when handling certificates? No. - IKEv2 tests have all been tweaked to use bigger timeouts (--impair suppress-retransmits or retransmit-timeout=1500) I guess the IKEv1 tests should be tweaked. I'll re-enable electric fence so we can see the progress. Andrew _______________________________________________ Swan-dev mailing list Swan-dev@lists.libreswan.org https://lists.libreswan.org/mailman/listinfo/swan-dev