6.2 becomes unresponsive under high traffic

[Kris Kennaway]

On Sat, Oct 21, 2006 at 10:06:16AM -0400, Mikhail Teterin wrote:
> Hello!
> 
> The system is a dual Opteron 244 running today's FreeBSD-6.2/amd64.
> 
> em-interface connects it to the switch (in gigabit mode).
> 
> When I direct 2 database dumps at the machine in parallel (the arriving data 
> is getting compressed and written to local disk), the "system" component of 
> the load (as reported by systat and top) goes up to 99-100% and stays there 
> for many minutes at a time. Accessing the box via console remains speedy, but 
> remote connections stall for minutes during which the box is not even 
> pingable...
> 
> What appears to wake it up, though, is hitting a (local) keyboard button...
> 
> Switching em0 to polling mode did not help...
> 
> "netstat -m" does not show any rejections of buffer requests.
> 
> It uses the BSD4-scheduler, as is the default.
> 
> Earlier, in the single-CPU configuration, the box had no problems dealing with 
> such 2 data streams for hours, backing up all our databases. We added another 
> processor and updated the world/kernel from 6.1 to 6.2, hoping to halve the 
> dump times -- and it is barely crowling now...
> 
> Please, advise. Thanks!

We've been discussing em issues for several weeks now, so it would be
great if you could get yourself up to speed - please review the
discussion on freebsd-stable and freebsd-net (start with posts by
Scott Long, myself, and Jack Vogel).

Kris
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-net/attachments/20061021/67bbab78/attachment.pgp