xfs
[Top] [All Lists]

Re: Parallel read performance crawls

To: KELEMEN Peter <fuji@xxxxxxx>
Subject: Re: Parallel read performance crawls
From: Rajagopal Ananthanarayanan <ananth@xxxxxxx>
Date: Thu, 24 May 2001 08:54:24 -0700
Cc: linux-xfs@xxxxxxxxxxx
References: <20010524162005.A26407@chiara.elte.hu>
Sender: owner-linux-xfs@xxxxxxxxxxx
KELEMEN Peter wrote:
> 
> Dear XFS developers,
> 
> While a single bonnie++ benchmarks shows excellent[1] performance
> on my test machine, two parallel bonnie++s still excel in write
> performance, but crawl[2] in read performance.  What is the effect
> I'm seeing here?

FWIW, in the following, I see not only read performance drop with 2 bonnies,
but also re-write. Are you using the same disk as the target for
the 2 bonnies? If so, I would strongly suspect disk head contention.
Delayed allocation/writes in the block output case seem to have a positive
effect: clusters of writes are fairly large; there is no such clustering
at the higher level for reads (re-write is read+write iirc).

Can you target 2 different disks for the 2 bonnies?

ananth.

> 
> Peter
> 
> [1]
> Version 1.01b       ------Sequential Output------ --Sequential Input- 
> --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec 
> %CP
> xfs_1            3G 11236  99 63619  51 17625  18 10133  88 58129  28  94.9   > 0
>                     ------Sequential Create------ --------Random 
> Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
> -Delete--
> files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
> %CP
>     16:15000:12/512   785  18 25534 101  1378  29  1582  37 24665  99  1575  
> 29
> 
> [2]
> Version 1.01b       ------Sequential Output------ --Sequential Input- 
> --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec 
> %CP
> xfs_2_1          2G 10851  98 42060  38  5277   5  4261  37  4040   1  47.8   > 0
>                     ------Sequential Create------ --------Random 
> Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
> -Delete--
> files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
> %CP
>     16:15000:12/512   313  10 22666  98   763  24   403  15 23511  99   182   
> 7
> 
> Version 1.01b       ------Sequential Output------ --Sequential Input- 
> --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec 
> %CP
> xfs_2_2          2G 10620  96 42498  39  5618   5  4296  37  3987   1  47.9   > 0
>                     ------Sequential Create------ --------Random 
> Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
> -Delete--
> files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
> %CP
>     16:15000:12/512   307   9 20354  99   804  22   428  14 23531 100   176   
> 6
> 
> --
>     .+'''+.         .+'''+.         .+'''+.         .+'''+.         .+''
>  Kelemen Péter     /       \       /       \       /      fuji@xxxxxxx
> .+'         `+...+'         `+...+'         `+...+'         `+...+'

-- 
--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------

<Prev in Thread] Current Thread [Next in Thread>