[Glass] RAID5 configurations

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[Glass] RAID5 configurations

GLASS mailing list
Ciao,

        i have a system with md software RAID5  configurations where i load GLASS 3.1.0.6.

        My questions is :    there is a preferable  md chunk-size for run glass ?

        Thank,

                Dario

       
_______________________________________________
Glass mailing list
[hidden email]
http://lists.gemtalksystems.com/mailman/listinfo/glass
Reply | Threaded
Open this post in threaded view
|

Re: [Glass] RAID5 configurations

GLASS mailing list
Dario,

The extents are written in 16k byte chunks and the tranlogs are written
in variable size chunks up to 64k bytes ... I'm not completely familiar
with how that translates into chunk size, but this should help you
figure out the rest ...

Dale

On 02/11/2015 01:14 PM, Dario Trussardi via Glass wrote:

> Ciao,
>
> i have a system with md software RAID5  configurations where i load GLASS 3.1.0.6.
>
> My questions is :    there is a preferable  md chunk-size for run glass ?
>
> Thank,
>
> Dario
>
>
> _______________________________________________
> Glass mailing list
> [hidden email]
> http://lists.gemtalksystems.com/mailman/listinfo/glass

_______________________________________________
Glass mailing list
[hidden email]
http://lists.gemtalksystems.com/mailman/listinfo/glass
Reply | Threaded
Open this post in threaded view
|

Re: [Glass] RAID5 configurations

GLASS mailing list
Ciao,

        i do some test for RAID5 performance.

        I have testing with chunk size of 512KB   and with chunk size of 8KB.


        I run the code:

  [ 1 to: 10000 do:[ :int | UserGlobals at: #Scandella put: (( UserGlobals at: #Scandella ) +1).
                                                        System commit ]] timeToRun.

        obtaining the same execution time.

        Is not strange ?

        With the chunk of 8KB i will expected better performance.

        Dario

> Dario,
>
> The extents are written in 16k byte chunks and the tranlogs are written in variable size chunks up to 64k bytes ... I'm not completely familiar with how that translates into chunk size, but this should help you figure out the rest ...
>
> Dale
>
> On 02/11/2015 01:14 PM, Dario Trussardi via Glass wrote:
>> Ciao,
>>
>> i have a system with md software RAID5  configurations where i load GLASS 3.1.0.6.
>>
>> My questions is :    there is a preferable  md chunk-size for run glass ?
>>
>> Thank,
>>
>> Dario
>>
>>
>> _______________________________________________
>> Glass mailing list
>> [hidden email]
>> http://lists.gemtalksystems.com/mailman/listinfo/glass
>
> _______________________________________________
> Glass mailing list
> [hidden email]
> http://lists.gemtalksystems.com/mailman/listinfo/glass

_______________________________________________
Glass mailing list
[hidden email]
http://lists.gemtalksystems.com/mailman/listinfo/glass
Reply | Threaded
Open this post in threaded view
|

Re: [Glass] RAID5 configurations

GLASS mailing list
Dario,

Performance testing for GemStone is a bit more complicated than what you
are doing here.

The number one disk-based performance issue is that both tranlogs and
extents are on the same LUN... Linux tends to prioritize reads over
writes and I have seen cases where tranlog writes can take up to a
second or more because of high volume read activity on the same LUN ...
Putting the extents and tranlogs on separate LUNs eliminates this
problem even thought the underlying disk may be the same...

But without looking at VSD you really cannot tell where your bottlenecks
are ...

Your test looks like a very straightforward test but the bottleneck
could be disk (high iowaits) or it could be cpu contention (cpus running
flat out) or ???

It is somewhat of an art for doing performance tuning for GemStone since
there are so many variables, in the past we have put on 1 week classes
on just performance tuning. There are also GemStone consultants who can
come in and do performance tuning...

If there is enough interest for a performance tuning class we might
consider putting on another one ... normally we travel to a site to give
the test (or have them travel to our offices) ... it might be possible
for us to put on a web-based class, but I'm not familiar with the
logistics of doing this...

Dale

On 2/19/15 9:42 AM, Dario Trussardi via Glass wrote:

> Ciao,
>
> i do some test for RAID5 performance.
>
> I have testing with chunk size of 512KB   and with chunk size of 8KB.
>
>
> I run the code:
>
>   [ 1 to: 10000 do:[ :int | UserGlobals at: #Scandella put: (( UserGlobals at: #Scandella ) +1).
> System commit ]] timeToRun.
>
> obtaining the same execution time.
>
> Is not strange ?
>
> With the chunk of 8KB i will expected better performance.
>
> Dario
>
>> Dario,
>>
>> The extents are written in 16k byte chunks and the tranlogs are written in variable size chunks up to 64k bytes ... I'm not completely familiar with how that translates into chunk size, but this should help you figure out the rest ...
>>
>> Dale
>>
>> On 02/11/2015 01:14 PM, Dario Trussardi via Glass wrote:
>>> Ciao,
>>>
>>> i have a system with md software RAID5  configurations where i load GLASS 3.1.0.6.
>>>
>>> My questions is :    there is a preferable  md chunk-size for run glass ?
>>>
>>> Thank,
>>>
>>> Dario
>>>
>>>
>>> _______________________________________________
>>> Glass mailing list
>>> [hidden email]
>>> http://lists.gemtalksystems.com/mailman/listinfo/glass
>> _______________________________________________
>> Glass mailing list
>> [hidden email]
>> http://lists.gemtalksystems.com/mailman/listinfo/glass
> _______________________________________________
> Glass mailing list
> [hidden email]
> http://lists.gemtalksystems.com/mailman/listinfo/glass

_______________________________________________
Glass mailing list
[hidden email]
http://lists.gemtalksystems.com/mailman/listinfo/glass