[ovs-discuss] How to diagnose ponctual slowlyness?
blp at nicira.com
Thu Dec 16 17:21:12 UTC 2010
On Thu, Dec 16, 2010 at 10:06:15AM +0100, Madko wrote:
> lookups: frags:0, hit:332572398, missed:39405270, lost:390
> Can you explain what means the "lookups" line, we don't find anything
> in the man pages, especially the "missed" and "lost" value. Thank you.
frags:0 means that no IP fragments were dropped
hit:332572398 means that 332572398 packets found matches when they were
looked up in the kernel flow table, so that they were processed entirely
in the kernel without involving userspace.
missed:39405270 means that 39405270 packets did not have matches in the
kernel flow table, so they had to be passed to userspace, and were.
That's a pretty high number. That likely indicates one of three things:
1. The switch has been up for a long time and there have been
many flows set up during that time.
2. There is some kind of traffic on the network that creates new
flows very quickly, e.g. high-rate port scans.
3. Some bug is keeping flows from being set up, and instead they
are being handled in userspace. (This could easily be the
cause of slowness.)
You should be able to tell whether #1 or #2 is the case on your own.
You can distinguish #3 by looking at how quickly "miss" increases when
your system is running at a normal speed versus how quickly it increases
when everything becomes very slow. If it goes up quickly only when
traffic is moving very slowly, then #3 is likely the problem.
lost:390 means that 390 packets that missed in the flow table were not
sent down to userspace because a kernel buffer overflowed.
> By the way we find a 1.0.3 version in the GIT repository, but no
> announce about it on the ovs website. Is this a stable release?
Hmm, I don't know why it wasn't announced. It is a stable release on
our lts-1.0 (long term support for version 1.0) branch.
More information about the discuss