resier4 performance with dbt-2

resier4 performance with dbt-2

[markw]

http://developer.osdl.org/markw/fs/dbt2_stp_results.html

I have some data (at the link above) with the reiser4 2004.02.06
snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
and 8-way systems.  On the 4-way systems, the results of the workload
seem comparable to other filesystems.  On the 8-way systems, the results
of the workload drop about 28%.

I have readprofile and oprofile (w/ annotated assembly) data with the
results.  If there's anything else I can provide let me know.

-- 
Mark Wong - - markw@osdl.org
Open Source Development Lab Inc - A non-profit corporation
12725 SW Millikan Way - Suite 400 - Beaverton, OR 97005
(503) 626-2455 x 32 (office)
(503) 626-2436      (fax)
http://developer.osdl.org/markw/

Re: resier4 performance with dbt-2

[Nikita Danilov]

markw@osdl.org writes:
 > http://developer.osdl.org/markw/fs/dbt2_stp_results.html
 > 
 > I have some data (at the link above) with the reiser4 2004.02.06
 > snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
 > and 8-way systems.  On the 4-way systems, the results of the workload
 > seem comparable to other filesystems.  On the 8-way systems, the results
 > of the workload drop about 28%.
 > 
 > I have readprofile and oprofile (w/ annotated assembly) data with the
 > results.  If there's anything else I can provide let me know.

That's very impressive. Thank you for data.

Do I understand correctly that you have applied the same reiser4 patch
(from 2004.02.06 snapshot) to the 2.6.2, 2.6.1-mm2, 2.6.1, and
2.6.0-mm2?

Also note strange results for ext2 on 4-way 2.6.1-mm2: it's like order
of magnitude slower than other contestants. Are test results averaged
over multiple runs?

 > 
 > -- 
 > Mark Wong - - markw@osdl.org

Nikita.

Re: resier4 performance with dbt-2

[markw]

On 13 Feb, Nikita Danilov wrote:
> markw@osdl.org writes:
>  > http://developer.osdl.org/markw/fs/dbt2_stp_results.html
>  > 
>  > I have some data (at the link above) with the reiser4 2004.02.06
>  > snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
>  > and 8-way systems.  On the 4-way systems, the results of the workload
>  > seem comparable to other filesystems.  On the 8-way systems, the results
>  > of the workload drop about 28%.
>  > 
>  > I have readprofile and oprofile (w/ annotated assembly) data with the
>  > results.  If there's anything else I can provide let me know.
> 
> That's very impressive. Thank you for data.
> 
> Do I understand correctly that you have applied the same reiser4 patch
> (from 2004.02.06 snapshot) to the 2.6.2, 2.6.1-mm2, 2.6.1, and
> 2.6.0-mm2?
> 
> Also note strange results for ext2 on 4-way 2.6.1-mm2: it's like order
> of magnitude slower than other contestants. Are test results averaged
> over multiple runs?

I've only been able to test reiser4 on 2.6.2.  I've been trying to track
filesystem performance and have done it with 2.6.2, 2.6.1-mm2, 2.6.1,
and 2.6.0-mm2, but I haven't been able to get results for every filesystem
for one reason or another.

Oh yeah, that ext2 4-way 2.6.1-mm2 result is way off.  In most cases, I
haven't tried repeating results.  But other than weird cases like that,
they usually don't vary so much.

Mark

Re: resier4 performance with dbt-2

[Hans Reiser]

Thanks Mark, you have most likely found a flaw in our code for us to fix 
next week.

Re: resier4 performance with dbt-2

[Dieter Nützel]

Am Freitag, 13. Februar 2004 18:31 schrieb markw@osdl.org:
> http://developer.osdl.org/markw/fs/dbt2_stp_results.html
>
> I have some data (at the link above) with the reiser4 2004.02.06
> snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
> and 8-way systems.  On the 4-way systems, the results of the workload
> seem comparable to other filesystems.  On the 8-way systems, the results
> of the workload drop about 28%.
>
> I have readprofile and oprofile (w/ annotated assembly) data with the
> results.  If there's anything else I can provide let me know.

Where are the ReiserFS 3.xx plus data-logging results?
Several sites have them in production.

Thanks,
	Dieter

Re: resier4 performance with dbt-2

[markw]

On 14 Feb, Dieter Nützel wrote:
> Am Freitag, 13. Februar 2004 18:31 schrieb markw@osdl.org:
>> http://developer.osdl.org/markw/fs/dbt2_stp_results.html
>>
>> I have some data (at the link above) with the reiser4 2004.02.06
>> snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
>> and 8-way systems.  On the 4-way systems, the results of the workload
>> seem comparable to other filesystems.  On the 8-way systems, the results
>> of the workload drop about 28%.
>>
>> I have readprofile and oprofile (w/ annotated assembly) data with the
>> results.  If there's anything else I can provide let me know.
> 
> Where are the ReiserFS 3.xx plus data-logging results?
> Several sites have them in production.

Unfortunitely we've been running into linux kernel problems that we
think is more of the fault of the aacraid driver on our systems than the
reiserfs 3.xx, so I don't have any results for that.  It's just odd that
only reiserfs 3.xx triggers it.  I'll copy that here for reference:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
 printing eip:                                                              
00000000      
*pde = 373ce001
*pte = 00000000
Oops: 0000 [#1]
CPU:    2      
EIP:    0060:[<00000000>]    Not tainted
EFLAGS: 00010046                        
EIP is at 0x0   
eax: f782e820   ebx: f7f9c000   ecx: f6f0c800   edx: c3afa6a0
esi: f782e800   edi: 00000086   ebp: 00000096   esp: f7f9ded0
ds: 007b   es: 007b   ss: 0068                               
Process swapper (pid: 0, threadinfo=f7f9c000 task=c38d2ca0)
Stack: f8919c7a c3afa6a0 f6fa1db0 f6f0c800 f782e9d0 f891df6d c3afa6a0 f6fa1db0 
       00000004 f782e800 f7f9c000 f7f9c000 f7f9c000 00000001 00000000 f6fbab88 
       f782e9d0 24000001 00000000 f7f9df80 f891e88b f7a9f180 c011c496 f71c8980 
Call Trace:                                                                    
 [<f8919c7a>] aac_io_done+0x3a/0x80 [aacraid]
 [<f891df6d>] aac_response_normal+0x18d/0x240 [aacraid]
 [<f891e88b>] aac_sa_intr+0x9b/0x120 [aacraid]         
 [<c011c496>] rebalance_tick+0x46/0xe0        
 [<c010bad9>] handle_IRQ_event+0x49/0x80
 [<c010becf>] do_IRQ+0xbf/0x1b0         
 [<c0106e70>] default_idle+0x0/0x40
 [<c0109f38>] common_interrupt+0x18/0x20
 [<c0106e70>] default_idle+0x0/0x40     
 [<c0106e9d>] default_idle+0x2d/0x40
 [<c0106f36>] cpu_idle+0x46/0x50    
 [<c0122768>] printk+0x178/0x1d0
 [<c040f0e6>] print_cpu_info+0x86/0xe0
                                      
Code:  Bad EIP value.
 <0>Kernel panic: Fatal exception in interrupt
In interrupt handler - not syncing

Re: resier4 performance with dbt-2

[Carl-Daniel Hailfinger]

markw@osdl.org wrote:
> On 14 Feb, Dieter Nützel wrote:
> 
>>Am Freitag, 13. Februar 2004 18:31 schrieb markw@osdl.org:
>>
>>>http://developer.osdl.org/markw/fs/dbt2_stp_results.html
>>>
>>>I have some data (at the link above) with the reiser4 2004.02.06
>>>snapshot applied to linux-2.6.2 with our DBT-2 (OLTP) workload on 4-way
>>>and 8-way systems.  On the 4-way systems, the results of the workload
>>>seem comparable to other filesystems.  On the 8-way systems, the results
>>>of the workload drop about 28%.
>>>
>>>I have readprofile and oprofile (w/ annotated assembly) data with the
>>>results.  If there's anything else I can provide let me know.
>>
>>Where are the ReiserFS 3.xx plus data-logging results?
>>Several sites have them in production.
> 
> 
> Unfortunitely we've been running into linux kernel problems that we
> think is more of the fault of the aacraid driver on our systems than the
> reiserfs 3.xx, so I don't have any results for that.  It's just odd that
> only reiserfs 3.xx triggers it.  I'll copy that here for reference:

I know several admins who will not use adaptec cards even if they receive
death threats. Regardless how high/low the load is, the machines will
eventually lock up when using adaptec SCSI controllers. No idea whether it
is the driver or the hardware, but replacing the controller with the same
model didn't fix it while exchanging it against another brand always fixed it.

Buggy hardware, buggy driver, I don't care. If it doesn't work out of the
box, another manufacturer gets the money next time.


Carl-Daniel