Re: bdflush/rpciod high CPU utilization, profile does not make sense

[Jakob Oestergaard <jakob@unthought.net>]

On Mon, Apr 11, 2005 at 11:21:45AM -0400, Trond Myklebust wrote:
> må den 11.04.2005 Klokka 16:41 (+0200) skreiv Jakob Oestergaard:
> 
> > > That can mean either that the server is dropping fragments, or that the
> > > client is dropping the replies. Can you generate a similar tcpdump on
> > > the server?
> > 
> > Certainly;  http://unthought.net/sparrow.dmp.bz2
> 
> So, it looks to me as if "sparrow" is indeed dropping packets (missed
> sequences). Is it running with NAPI enabled too?

Yes, as far as I know - the Broadcom Tigeon3 driver does not have the
option of enabling/disabling RX polling (if we agree that is what we're
talking about), but looking in tg3.c it seems that it *always*
unconditionally uses NAPI...

sparrow:~# ifconfig
eth0      Link encap:Ethernet  HWaddr 00:09:3D:10:BB:1E
          inet addr:10.0.1.20  Bcast:10.0.1.255  Mask:255.255.255.0
          inet6 addr: fe80::209:3dff:fe10:bb1e/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:2304578 errors:0 dropped:0 overruns:0 frame:0
          TX packets:2330829 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:2381644307 (2.2 GiB)  TX bytes:2191756317 (2.0 GiB)
          Interrupt:169

No dropped packets... I wonder if the tg3 driver is being completely
honest about this...

Still, 2.4 manages to perform twice as fast against the same server.

And, the 2.6 client still has extremely heavy CPU usage (from rpciod
mainly, which doesn't show up in profiles)

The plot thickens...

Trond (or anyone else feeling they might have some insight they would
like to share on this one), I'll do anything you say (ok, *almost*
anything you say) - any ideas?


-- 

 / jakob