[strongSwan] Out of control software interrupts

jsullivan at opensourcedevel.com jsullivan at opensourcedevel.com
Sat Jun 6 04:15:34 CEST 2015


Hello, all.  We are still trying to get to the bottom of our very poor
performance on GbE connections with high (~=80ms) latency.  We have noticed that
on the production gateways (as opposed to test), software interrupts are through
the roof.  The major traffic was only 60 Mbps of UDP video to one site.  Packet
loss was horrendous even on barely utilized links and si was 100% on a single
processor.  pcrypt does not seem to help with the interrupts.
 
We installed irqbalance and that helped quite a bit.  We also played with
 /proc/sys/net/ipv4/xfrm4_gc_thresh and found an optimal setting at 2048000 but
that only brought our end to end nuttcp throughput from 4 Mbps to 40 Mbps and
eliminated the packet loss.  We should be able to get at least 400 Mbps and I
would really expect closer to 800 Mbps.  We have proved we can exceed that on a
single thread using UDP.
 
What else can we do to get these software interrupts under control?
 
kernel is 3.2.0-4-amd64 #1 SMP Debian 3.2.65-1+deb7u1 x86_64 GNU/Linux
 
Thanks - John
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.strongswan.org/pipermail/users/attachments/20150605/478aabdc/attachment.html>


More information about the Users mailing list