Veritas-bu

Re: [Veritas-bu] Tuning backups.

2007-10-12 20:15:41
Subject: Re: [Veritas-bu] Tuning backups.
From: "Scott Jacobson" <SJACOBSO AT novell DOT com>
To: "Justin Piszcz" <jpiszcz AT lucidpixels DOT com>, "Andrew Stueve" <andrew.stueve AT neovera DOT com>
Date: Fri, 12 Oct 2007 18:04:26 -0600
For Linux,
 
(LTO II)
 
In the /usr/src/linux-2.6.5-7.282/drivers/scsi location, I've modified (as per the Veritas performance options) the st_options.h file.
 
default:
#define ST_FIXED_BUFFER_BLOCKS 32
new:
#define ST_FIXED_BUFFER_BLOCKS 64
 
to match the default 64K default of BPTM.
 
I've seen maybe a 2% or 3% increase in performance, not much, but I'll take anything I can get.
 
-sj


>>> Justin Piszcz <jpiszcz AT lucidpixels DOT com> 10/12/2007 4:48 PM >>>
I have and there are more options too you can set, as long as you set the
NUM_BUFFERS and SIZE_BUFFERS, those give optimal results, I was not able
to increase the bandwidth to the (LTO-2) tapes I was testing using any of
the st parameters.

Justin.

On Fri, 12 Oct 2007, Andrew Stueve wrote:

> Was there any resolution here?  Anyone test this by adding to the boot
> param line?
>
> Adding something like:  st=buffer_kbs:256,max_buffers:6
>
> Anyone? Beuhler?
>
> -Andrew
>
> Justin Piszcz wrote:
>>
>> On Wed, 5 Sep 2007, Justin Piszcz wrote:
>>
>>>
>>> On Wed, 5 Sep 2007, Parker, Matthew wrote:
>>>
>>>> What type of benchmark testing are you looking for?
>>>>
>>>> -----Original Message-----
>>>> From: veritas-bu-bounces AT mailman.eng.auburn DOT edu [mailto:veritas-bu-bounces AT mailman.eng.auburn DOT edu] On Behalf Of Justin Piszcz
>>>> Sent: Wednesday, September 05, 2007 4:36 AM
>>>> To: veritas-bu AT mailman.eng.auburn DOT edu
>>>> Subject: [Veritas-bu] Tuning backups.
>>>>
>>>> There are several ways to tune backups.
>>>>
>>>> 1. NET_BUFFER_SZ
>>>> 2. SIZE_DATA_BUFFERS
>>>> 3. NUM_DATA_BUFFERS
>>>>
>>>> While these help a little, I found by reading the NBU6 optimization document I improved speed by 2x in some cases (need to do some more
>>>> verification) using 256k buffers in the kernel (Linux) st driver rather than the default of 32k and also using more of them.
>>>>
>>>> Anyone here do a LOT of benchmark testing?
>>> So far, it would appear that using general optimizations in NetBackup
>>> only help so much, such as the ones mentioned above with the exception
>>> of NET_BUFFER_SZ, I still need to look into that one.
>>>
>>> The buffers in the kernel appear to help much more, in Linux 2.4 the
>>> default buffers for the st tape driver are 32k and there's four of them.
>>>
>>> I want to know every possible tunable parameter and its effect on the
>>> system for the fastest backup possible.
>>>
>>> Justin.
>>>
>>> _______________________________________________
>>> Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
>>> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>>>
>>
>> In particular:
>>
>> 9.3.1. st boot parameters
>>
>> st=xxx[,yyy] where xxx is one of the following:
>> buffer_kbs:<n>
>> write_threshold_kbs:<n>
>> max_buffers:<n>
>> max_sg_segs:<n>
>>
>> (The old boot parameters st=aa[,bb[,cc[,dd]]] supported but deprecated)
>>
>>
>> http://www.linux.com/base/ldp/howto/SCSI-2.4-HOWTO/st.html
>> _______________________________________________
>> Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
>> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>>
> _______________________________________________
> Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>
_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
<Prev in Thread] Current Thread [Next in Thread>