Quantcast

Does iometer-2004.07.30 work properly on Redhat 2.6 kernels?

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Does iometer-2004.07.30 work properly on Redhat 2.6 kernels?

George Lestat
Running iometer under 2.4 kernels seems to produce better results than on 2.6.
 
 
I have a test I created called OLTP2K that does:
Transfer Request Size: 2K
100% of size
67% reads
100% random
0 delay
1 burst
0 align
0 reply
 
QDepth 1-64
Raw Disk
 
Using the OLTP2K test on 2.4 kernels produces performance results that increase with larger QDepths. Running these same tests on a 2.6 kernel produces the same result for all QDepths. And the QDepth does not appear to actually grow beyond 1.
 
The Supported platforms page only mentions 2.4 kernels. I get the same results using the development version.
 
Could there be a problem with the AIO libraries?
 
Also, the results for these tests are inflated unless I add the O_DIRECT flag to the open() call.
 
Has anyone else seen this problem?
 
 
Thanks,
Don
 
 
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Does iometer-2004.07.30 work properly on Redhat 2.6 kernels?

Ming Zhang

On Tue, 2005-10-11 at 17:05 -0500, George Lestat wrote:

> Running iometer under 2.4 kernels seems to produce better results than
> on 2.6.
>  
>  
> I have a test I created called OLTP2K that does:
>         Transfer Request Size: 2K
>         100% of size
>         67% reads
>         100% random
>         0 delay
>         1 burst
>         0 align
>         0 reply
>          
>         QDepth 1-64
>         Raw Disk
>          
> Using the OLTP2K test on 2.4 kernels produces performance results that
> increase with larger QDepths. Running these same tests on a 2.6 kernel
> produces the same result for all QDepths. And the QDepth does not
> appear to actually grow beyond 1.
>  
> The Supported platforms page only mentions 2.4 kernels. I get the same
> results using the development version.
>  
> Could there be a problem with the AIO libraries?

it is possible. aio changed a lot sine 2.4 to 2.6.

>  
> Also, the results for these tests are inflated unless I add the
> O_DIRECT flag to the open() call.

what u mean inflated?

>  
> Has anyone else seen this problem?
>  
>  
> Thanks,
> Don
>  
>  



-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
Iometer-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/iometer-user
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Does iometer-2004.07.30 work properly on Redhat 2.6 kernels?

Ming Zhang
it is interesting since dynamo are working on raw disk and i assumed u
raw disk size is much larger than u ram size, so cache effect should not
have that big impact. u might also need some ram up time and longer
running time.

ming



On Thu, 2005-10-13 at 13:05 -0500, Don Brace wrote:

> The results are higher than they should be. I think that this may be
> due to the buffer cache.
>  
> O_DIRECT flag takes out the cache, I think.
>  
> Thanks for the reply.
>
>  
> On 10/13/05, Ming Zhang <[hidden email]> wrote:
>        
>         On Tue, 2005-10-11 at 17:05 -0500, George Lestat wrote:
>         > Running iometer under 2.4 kernels seems to produce better
>         results than
>         > on 2.6.
>         >
>         >
>         > I have a test I created called OLTP2K that does:
>         >         Transfer Request Size: 2K
>         >         100% of size
>         >         67% reads
>         >         100% random
>         >         0 delay
>         >         1 burst
>         >         0 align
>         >         0 reply
>         >
>         >         QDepth 1-64
>         >         Raw Disk
>         >
>         > Using the OLTP2K test on 2.4 kernels produces performance
>         results that
>         > increase with larger QDepths. Running these same tests on a
>         2.6 kernel
>         > produces the same result for all QDepths. And the QDepth
>         does not
>         > appear to actually grow beyond 1.
>         >
>         > The Supported platforms page only mentions 2.4 kernels. I
>         get the same
>         > results using the development version.
>         >
>         > Could there be a problem with the AIO libraries?
>        
>         it is possible. aio changed a lot sine 2.4 to 2.6.
>        
>         >
>         > Also, the results for these tests are inflated unless I add
>         the
>         > O_DIRECT flag to the open() call.
>        
>         what u mean inflated?
>        
>         >
>         > Has anyone else seen this problem?
>         >
>         >
>         > Thanks,
>         > Don
>         >
>         >
>        
>



-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
Iometer-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/iometer-user
Loading...