Re: FreeBSD 5.3b7and poor ata performance

From: Scott Long <scottl_at_freebsd.org>
Date: Mon, 18 Oct 2004 23:23:23 -0600
fandino wrote:
> Hello Søren,
> 
> Søren Schmidt wrote:
> 
>>>> GNU/Linux 2.4.18 with ext2:               56848 K/sec
>>>> FreeBSD 5.3b7 with default fs:            26347 K/sec
>>>> FreeBSD 5.3b7 with default fs(async):     26566 K/sec
>>>> FreeBSD 5.3b7 ata raid0* (two disks):     26131 K/sec
>>>> FreeBSD 5.3b7 geom stripe* (two disks):   30063 K/sec
>>>> FreeBSD 5.3b7 geom stripe** (four disks): 31891 K/sec
>>>> OpenBSD 3.5 UFS fs:                       55277 K/sec
>>>>
>>>> * Each disk of the raid had a throughput of approx. 15000 K/sec
>>>> ** Each disk of the raid had a throughput of approx. 7500 K/sec
>>>> Each disk of the read split the throughput by half.
>>>>
>>>> How is possible that FreeBSD performs as bad?
>>>>
>>>>
>>> If you're still using the GENERIC kernel, that could explain it, and 
>>> judging
>>> from other emails I've seen from you, you're still using the GENERIC 
>>> kernel.
>>
>>
>> Right, and you should also use -U (softupdates) on you newfs line.
> 
> 
> 
> 
> FreeBSD 5.3b7 with default fs(+softupdates):   26615 K/sec
> http://195.55.55.164/tests/fbsd+softupdates.txt
> 
> 
> 
> I was running FreeBSD-4.x for two years with this problem, waiting
> for FreeBSD-5 because ATAng looks very promising.
> 
> Unfortunately the performance problem persist :-( and I'd like to
> call the attention about performance over raw devices, whilst it's
> a very scientific test it's very curious:
> 
> # dd if=/dev/ad4 of=/dev/null bs=1024k count=1024
> 1024+0 records in
> 1024+0 records out
> 1073741824 bytes transferred in 31.090536 secs (34535970 bytes/sec)
> 
> over 34000 K/sec, using raw devices (for sequential access obviously)
> not softupdates, filesystems or caches are involved, and with all this
> FreeBSD performace is very deficient. Tests with OpenBSD and Linux
> using raw devices shows a throughput of approx 60000 K/sec.
> The question here is why using low-level access to disks is so bad?
> 
> Perphas I'm missing something but this seems very weird to me.
> 
> I'd like to know wich is you opinion about this.
> 
> Thank you.

After you run your test, can you send the output of 'sysctl hw.busdma'?
Thanks,

Scott
Received on Tue Oct 19 2004 - 03:24:55 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:18 UTC