If you care about, I have some profile info with Netbench on various Linux
kernel versions, as well as some SPEC SFS profiles.
d_lookup, lookup_dentry were the ones that got relatively high CPU counts.
Here is one with pre5 kernel runing on
a 4-way Dell poweredge with one Alteon gigabit card and 60 clients.
Besides the potential performance problems, I also saw lots of error
messages with Netbench tests. The Netbench errors
seem to have a fixed pattern. That is, the errors mostly occur on
Netbench's file rename and delete operations. Unfortunately,
I don't have access to the Netbench source to check on what kind of errors
they are. But this does make me feel suspicious about whether
there is something happened with the file system on Linux. I tried the same
thing with pre3 and 2.3.51, I saw more of
this kind of errors with them than with pre5. I have not tried even
earlier versions though. There is nothing wrong happened on the Linux
server side. So, I don't really have a good idea as to what caused the
problem, except for some guessing.
Also, I don't know if the problem would repeat with other benchmarks, like
SPEC SFS. I have not tried myself.
(See attached file: jbods_pre5_c60)
Here is a SPEC SFS profile with 2.3.35 running on a 2-way Intellistation.
Sorry, I don't have any newer profiles with SPEC SFS.
But my guess is that the profile would not change much. At least this has
been the case with Netbench.
(See attached file: specsfs3100iops_linux2335smp_nosync_nfs2_18c.prof)
Hope this would be useful.
Ying
Alexander Viro <viro@math.psu.edu>@vger.rutgers.edu on 04/18/2000 01:13:02
AM
Sent by: owner-linux-kernel@vger.rutgers.edu
To: Hans Reiser <hans@reiser.to>
cc: "Jeff V. Merkey" <jmerkey@timpanogas.com>,
"linux-kernel@vger.rutgers.edu" <linux-kernel@vger.rutgers.edu>,
"yura@namesys.botik.ru" <yura@namesys.botik.ru>
Subject: Re: Are you all aware that there was a slight drop in performance
in going from pre3 to pre5?
On Tue, 18 Apr 2000, Hans Reiser wrote:
> Thanks Jeff, anyone on the list know what code/author is the reason why
Jeff is
> seeing more calls to d_lookup?
It's not necessarily bad. I'ld love to look at the whole picture - right
now the whole lookup mechanism is in the middle of a sequence of changes
and timings definitely _will_ change several times before it will settle
down. Jeff, care to email the results of profiling you've got?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 23 2000 - 21:00:13 EST