Google
  Web www.spinics.net

Re: Ping RTT clock resolution problem on x86_64

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


Not yet, is it open to the public? We have some indication that the "tsc" kernel flag fixes some of the timer problems.

-Federico

On Oct 22, 2004, at 7:27 AM, Joshua Jensen wrote:

Please say that you have entered this in bugzilla... this is important

Joshua



On Thu, Oct 14, 2004 at 01:58:57PM -0700, Federico Sacerdoti wrote:
Hello,

I am a developer from the Rocks project in San Diego. We are noticing a
problem with the clock resolution for TCP round trip times that only
emerges on AMD Opteron systems. The bug is characterized as follows:


The issue is that the clock resolution is only 10ms: ie any rtt value
between 0-10ms is reported as 0.00ms.

--On Opteron (bug)--
root@www root]# ping google.com
PING google.com (216.239.37.99) 56(84) bytes of data.
64 bytes from 216.239.37.99: icmp_seq=0 ttl=238 time=89.9 ms
64 bytes from 216.239.37.99: icmp_seq=1 ttl=238 time=79.9 ms
64 bytes from 216.239.37.99: icmp_seq=2 ttl=236 time=79.9 ms

[root@www root]# ping nbc.com
PING nbc.com (63.241.60.70) 56(84) bytes of data.
64 bytes from nbc.com (63.241.60.70): icmp_seq=0 ttl=53 time=0.000 ms
64 bytes from nbc.com (63.241.60.70): icmp_seq=1 ttl=53 time=0.000 ms


--On Other systems (normal)-- [root@rocks-43 root]# ping google.com PING google.com (216.239.57.99) 56(84) bytes of data. 64 bytes from 216.239.57.99: icmp_seq=0 ttl=244 time=11.5 ms 64 bytes from 216.239.57.99: icmp_seq=1 ttl=244 time=11.8 ms 64 bytes from 216.239.57.99: icmp_seq=2 ttl=244 time=11.5 ms 64 bytes from 216.239.57.99: icmp_seq=3 ttl=244 time=11.7 ms

[root@rocks-43 root]# ping nbc.com
PING nbc.com (63.241.60.70) 56(84) bytes of data.
64 bytes from nbc.com (63.241.60.70): icmp_seq=0 ttl=53 time=7.41 ms
64 bytes from nbc.com (63.241.60.70): icmp_seq=1 ttl=53 time=7.32 ms
64 bytes from nbc.com (63.241.60.70): icmp_seq=2 ttl=53 time=7.39 ms
64 bytes from nbc.com (63.241.60.70): icmp_seq=3 ttl=53 time=7.33 ms


This behavior happens for several different types of ethernet cards on opterons, and is not observed on either Pentium or Nocona systems.

Has anyone noticed this before? Our kernel version is:

(Nocona)
Linux rocks-43.sdsc.edu 2.4.21-20.EL #1 SMP Fri Oct 1 09:12:23 PDT 2004
x86_64 x86_64 x86_64 GNU/Linux


(Opteron)
Linux www.rocksclusters.org 2.4.21-20.ELsmp #1 SMP Sat Sep 18 18:28:16
PDT 2004 x86_64 x86_64 x86_64 GNU/Linux

Thank you,
Federico

Rocks Cluster Group, San Diego Supercomputer Center, CA

--

amd64-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/amd64-list

--
Joshua Jensen
joshua@xxxxxxxx
"If God didn't want us to eat animals, why did he make them out of meat?"


--

amd64-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/amd64-list

Federico

Rocks Cluster Group, San Diego Supercomputer Center, CA

--

amd64-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/amd64-list

[Search]     [Home]     [Kernel List]     [Linux ia64]     [Linux X86_64]     [Red Hat Install]     [Red Hat Migration]     [Red Hat Development]     [Red Hat 9 Bible]     [Red Hat 9 Mailing List]     [Fedora Legacy]     [Yosemite News]