Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 26 Oct 2009 14:10:22 +0100
From:      Arnaud Houdelette <arnaud.houdelette@tzim.net>
To:        Jaime Bozza <jbozza@mindsites.com>
Cc:        "freebsd-stable@freebsd.org" <freebsd-stable@freebsd.org>, Jacob Myers <jacob@whotookspaz.org>
Subject:   Re: Possible scheduler (SCHED_ULE) bug?
Message-ID:  <4AE59FBE.6060904@tzim.net>
In-Reply-To: <CF092998EA82AE4D83AF9C9ECA9A8BDE02EB411A8F@ATLAS.msg.local>
References:  <CF092998EA82AE4D83AF9C9ECA9A8BDE02EB411A85@ATLAS.msg.local>	<d5992baf0910231149qa4e89fdx2e1cde9b2a529cd3@mail.gmail.com>	<CF092998EA82AE4D83AF9C9ECA9A8BDE02EB411A8E@ATLAS.msg.local>	<4AE2232E.10406@whotookspaz.org> <CF092998EA82AE4D83AF9C9ECA9A8BDE02EB411A8F@ATLAS.msg.local>

next in thread | previous in thread | raw e-mail | index | archive | help
Jaime Bozza a écrit :
>>> The additional information I have (over the PR) is that:
>>> 1) Files over 64K cause the problem, not just larger files
>>>       
>> I thought it was over 1 MB or so. But maybe I'm wrong. ISTR that I
>> couldn't trigger it with some images of around 70K.
>>     
>
> I discovered it originally with a 72K file.  After some tests, I found a 63K file worked and a 65K file didn't.  When I get back into the office, I can test the actual boundary (65535, 65536, 65537, etc), but 64K seems pretty logical.
>
>   
>>> 2) switching over to SCHED_4BSD eliminates the problem - system no
>>>       
>> longer locks.
>> I will have to test this. This is indeed interesting...
>>
>>     
>>> 3) 7.2 amd64 doesn't have the problem - Tested a similar
>>>       
>> configuration and was not able to duplicate on amd64 at all.
>> I can replicate this problem on FreeBSD 7.2/amd64 reliably.
>>     
>
> I haven't tried larger files - Maybe the boundary is different on amd64?   Doing some quick tests right now, I was able to upload a 100MB file without a problem, but this is an AMD64 system with SMP, plus the filesystem is all ZFS, so there are too many things different.  I'll have to setup a system that closely mirrors the rest of my tests (UFS, ULE, no SMP, etc) before I can say I'm not having a problem there.
>
> Jaime
>   
I had the same issue using 7.1 amd64, with ZFS, no SMP.
Not really sure what is the size boundary. I can't really test either, 
as the machine is remote.
But I confirm that each tentative upload of certain relatively 'big' 
files (around 1MB) with wordpress hanged the system before I switched 
from sendfile to writev.

I might do some test on amd64 7.2 with no SMP if it can be of any use ?

Arnaud



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4AE59FBE.6060904>