Network performance issues when writing to disk (5.2.1-RELEASE)

[Robert Watson]

On Tue, 10 Aug 2004, Erik Rothwell wrote:

> I have a workgroup server running FreeBSD 5.2.1 that is experiencing
> some serious network performance problems over very minimal load.

A couple of background questions:

- What sort of kernel configuration are you using -- GENERIC, or a custom
  kernel, and if so, is it an SMP box and is SMP enabled in the kernel?

- When the system performance suddenly degrades, what is the apparent load
  and condition of the system?  In particular, at that point if you
  measure the load average and CPU utilization, perhaps with "systat
  -vmstat" or "top", are the CPU(s) maxed out?  How much time is in user
  vs. system vs. interrupt (vs. idle).

- Can you confirm that the system does or doesn't have enough memory, and
  is or isn't paging much when the degredation occurs?  "systat -vmstat"
  will probably be useeful here also.

> The server has two miibus-based NICs: a WAN link via dc1 and a switched
> LAN serving ~7 clients attached via dc0. 
> 
> File transfers to the server (eg, scp, WebDAV, NFS, FTP) experience
> terrible throughput, starting off at about 2MB/s and dropping rapidly to
> ~200KB/s.

It would be interesting to know what condition the processes in question
are in.  In particular, how much time they're spending blocked in network,
how much in CPU, and how much I/O bound to/from disk.  It might be useful
to ktrace the process, although ktrace will change the system load by
consuming some combination of memory, disk, and cpu, of course.

> At first, I suspected duplex mismatch or a misbehaving switch (unmanaged
> switch with all client interfaces set to autosense), but the problem can
> be replicated with the server connected directly to any of the clients
> with the media at both ends explicitly set. netstat -ni indicates no
> collisions but a few I/Oerrs during transfers. 
> 
> It appears from testing that the interface itself responds
> intermittently under this load (during transfers, the switch reports the
> port is intermittently inactive -- ifconfig on the server will also
> sometimes report "status: inactive") and packet loss approaches 75%.
> Occasionally, "dc0: TX underrun -- increasing TX threshold" would crop
> up in dmesg, but that's not usually noteworthy.

This is the if_dc interface realizing that the packet flow is going
faster and accomodating.  However, the intermittent response thing sounds
like a more serious problem.  It might be useful to run a long-running
ping session using "ping -c" over a couple of hours and monitor the
connectivity of the box from another box and see if it comes and goes.  A
few years ago I ran into some weird problems with if_dc cards interacting
poorly with several linksys switches.  I found that they saw intermittent
connectivity, and autonegotiation worked poorly between them and that
particular line of switches.  It could be that you're experiencing
something similar.

> I replaced the NIC, yet the problem curiously persists. 

Did you replace it with another if_dc card, or with a card using a
different interface driver?

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research