SUMMARY: Dell GX260 / x86 Sol 9 / NIC very slow

From: Geoff Lane <zzassgl_at_zoe.mcc.ac.uk>
Date: Wed Jun 02 2004 - 07:14:12 EDT
Hidden in the tidal wave of "out of office" replies were a couple of
suggestions to look at NIC autonegotiation.  I've checked that and even when
the NIC is forced to the required 100M/duplex the problems remain.  Plus
that doesn't explain the constant 50% kernel CPU.

By applying patches by hand I'm now certain that the problem is caused by
something in 112234-12 (112234-11 doesn't show the symptoms.)

The only obvious references to networking I can see in -12 are

4697693 Restore tcp MSS negotiation/adjustment that occurred in Solaris
	2.6/2.5.1
4705144 IPV4 and IPV6 behave differently with addresses assigned to loopback

Unfortunately 4697693 doesn't link to a description on Sunsolve so that's a
dead end.

So, I don't have a solution yet (apart from removing 112234-12 which I've
done) but I suspect any solution will be related to 4697693.

Thanks for the replies.

---------------------------------------------------------------------------


On Tue, Jun 01, 2004 at 03:51:27PM +0100, Geoff Lane wrote:
> I'm running Solaris 9 on a Dell GX260 with Intel NIC chips on the motherboard
> (equiv to an Intel Pro/1000 card and using the latest INTCGigaE drivers from
> Intel.)
> 
> The current recommended patch set makes the PC act in a very strange manner.
> 
> The kernel CPU goes to about 50% and stays there and network connections are
> VERY slow. If you ping the system from another machine you can watch the
> ping times slowly increment from sub-millisecond response to a 1 second
> response in 10 or 20 mSec jumps.  When it hits 1 second the time starts from
> a very low value again. There is also a 5% packet loss.  Normally pings on
> this network will run in sub-millisecond times and zero packet loss.
> 
> 64 bytes from zoe : icmp_seq=3181 ttl=255 time=964 ms
> 64 bytes from zoe : icmp_seq=3182 ttl=255 time=978 ms
> 64 bytes from zoe : icmp_seq=3183 ttl=255 time=981 ms
> 64 bytes from zoe : icmp_seq=3184 ttl=255 time=995 ms
> 64 bytes from zoe : icmp_seq=3185 ttl=255 time=1008 ms
> 64 bytes from zoe : icmp_seq=3186 ttl=255 time=9.15 ms
> 64 bytes from zoe : icmp_seq=3187 ttl=255 time=22.7 ms
> 64 bytes from zoe : icmp_seq=3188 ttl=255 time=26.3 ms
> 64 bytes from zoe : icmp_seq=3189 ttl=255 time=40.1 ms
> 64 bytes from zoe : icmp_seq=3190 ttl=255 time=53.8 ms
> 64 bytes from zoe : icmp_seq=3191 ttl=255 time=67.3 ms
> 64 bytes from zoe : icmp_seq=3192 ttl=255 time=80.8 ms
> 64 bytes from zoe : icmp_seq=3193 ttl=255 time=90.1 ms
> 64 bytes from zoe : icmp_seq=3194 ttl=255 time=100 ms
> 64 bytes from zoe : icmp_seq=3195 ttl=255 time=120 ms
> 64 bytes from zoe : icmp_seq=3196 ttl=255 time=133 ms
> 64 bytes from zoe : icmp_seq=3197 ttl=255 time=1172 ms
> 64 bytes from zoe : icmp_seq=3198 ttl=255 time=1184 ms
> 
> I've tracked the problem down to something in the 112234 series of Kernel
> patches.  The 112234-08 version works OK, the 112234-12 (and newest) version
> shows the problem.
> 
> Has anybody already worked out what is happening here?
> 
> Thanks, summary will follow.
> 
> -- 
> | Geoff. Lane | Manchester Computing | Manchester | M13 9PL | England |
> 
> "Bother", said Pooh, as he sunk his twelfth Guinness
> _______________________________________________
> sunmanagers mailing list
> sunmanagers@sunmanagers.org
> http://www.sunmanagers.org/mailman/listinfo/sunmanagers

-- 
| Geoff. Lane | Manchester Computing | Manchester | M13 9PL | England |

"Bother", said Pooh, as he kicked hell out of his modem
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Wed Jun 2 07:14:05 2004

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:32 EST