6:32 p.m.
Also, your journal must be pathetically small, maybe only 4*500 blocks, or 8MB, which is
almost the minimum size. This is not how mkfs.lustre will format it normally.
I think there is a section on the manual, or maybe the tune2fs man page on how to change
the journal size. It is done in the same manner as ext3 and ext4.
Cheers, Andreas
On 2013-03-06, at 23:27, "Niu, Yawei" <yawei.niu(a)intel.com> wrote:
Hi, Andersen
The error message is saying that: a write request requires too many
journal credits, it even exceeded the maximum credits for a single
transaction (500), and we just turn to use the maximum value instead of
calculated one (555).
If there are lots write requests in parallel requires so much journal
credits, the system could be easily running short of journal space,
that'll cause performance drops since lot of requests have to wait on
journal space.
I guess the reason of write requiring so much credits is the write request
contains many noncontiguous segments spreading over the file, it probably
caused by some IO pattern changes of your application?
Thanks
- Niu
On 3/6/13 3:31 AM, "Andersen, Per" <per.andersen(a)ttu.edu> wrote:
> We are running lustre 1.8.8 on centos 5.8 for the servers and the clients
> are lustre 1.8.5 on centos 5.5. We have run trouble free with that setup
> for over a year, the errors we are seeing now appeared in the last few
> weeks and on only one OSS. I have goggled the error and found nothing
> enlightening.
>
> Per Andersen Ph.D. P.E.
> HPCC
> Texas Tech University
> 806-834-6924
>
>
>
> -----Original Message-----
> From: Jones, Peter A [mailto:peter.a.jones@intel.com]
> Sent: Tuesday, March 05, 2013 1:01 PM
> To: Andersen, Per; hpdd-discuss(a)lists.01.org
> Subject: Re: [HPDD-discuss] Lustre error
>
> Per
>
> It would probably be helpful to know which release of Lustre that you are
> running
>
> Peter
>
> On 3/5/13 6:25 AM, "Andersen, Per"
> <per.andersen@ttu.edu<mailto:per.andersen@ttu.edu>> wrote:
>
> We started getting huge numbers of these messages on one of our OSS
> servers, there are two OSS servers in the system each with 2 OSTs on
> them. The second OSS does not generate these messages.
>
> Mar 5 08:20:20 rt-oss1 kernel: LustreError:
> 29297:0:(fsfilt-ldiskfs.c:493:fsfilt_ldiskfs_brw_start()) want too many
> journal credits (555) using 500 instead Mar 5 08:20:20 rt-oss1 kernel:
> LustreError: 29297:0:(fsfilt-ldiskfs.c:493:fsfilt_ldiskfs_brw_start())
> Skipped 16215 previous similar messages
>
> Does anyone know what this message is about? Since we started getting
> them the lustre system performance has been reduced.
>
> Per Andersen Ph.D. P.E.
> HPCC
> Texas Tech University
> 806-834-6924
>
> _______________________________________________
> HPDD-discuss mailing list
> HPDD-discuss(a)lists.01.org
> https://lists.01.org/mailman/listinfo/hpdd-discuss
_______________________________________________
HPDD-discuss mailing list
HPDD-discuss(a)lists.01.org
https://lists.01.org/mailman/listinfo/hpdd-discuss