Two questions about stripe_cache_size

Two questions about stripe_cache_size

[Patrick J. LoPresti]

(Thanks to Piergiorgio Sartor for pointing out this setting to me.)

I am looking for documentation on /sys/block/mdX/md/stripe_cache_size.
 Specifically:

1) What are the units of this attribute?

2) I see it is supported for RAID 5.  Is it supported for RAID 6?

Thanks!

 - Pat

Re: Two questions about stripe_cache_size

[NeilBrown]

On Tue, 22 Feb 2011 13:46:57 -0800 "Patrick J. LoPresti" <lopresti@gmail.com>
wrote:

> (Thanks to Piergiorgio Sartor for pointing out this setting to me.)
> 
> I am looking for documentation on /sys/block/mdX/md/stripe_cache_size.
>  Specifically:
> 
> 1) What are the units of this attribute?

Pages per device (1 page is 4K normally)

> 
> 2) I see it is supported for RAID 5.  Is it supported for RAID 6?

Yes.

NeilBrown

Re: Two questions about stripe_cache_size

[Roberto Spadim]

could add ' pages' or ' kB' after number? just to know what´s unit

2011/2/22 NeilBrown <neilb@suse.de>:
> On Tue, 22 Feb 2011 13:46:57 -0800 "Patrick J. LoPresti" <lopresti@gmail.com>
> wrote:
>
>> (Thanks to Piergiorgio Sartor for pointing out this setting to me.)
>>
>> I am looking for documentation on /sys/block/mdX/md/stripe_cache_size.
>>  Specifically:
>>
>> 1) What are the units of this attribute?
>
> Pages per device (1 page is 4K normally)
>
>>
>> 2) I see it is supported for RAID 5.  Is it supported for RAID 6?
>
> Yes.
>
> NeilBrown
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>



-- 
Roberto Spadim
Spadim Technology / SPAEmpresarial
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Two questions about stripe_cache_size

[Patrick J. LoPresti]

Thanks, Neil.  Actually, I think I have one more question...  (Well,
one question with many faces.)

Is the stripe cache write-back, or write-through?  If the latter, when
does it get flushed to disk, exactly?

 - Pat

Re: Two questions about stripe_cache_size

[NeilBrown]

On Tue, 22 Feb 2011 14:13:14 -0800 "Patrick J. LoPresti" <lopresti@gmail.com>
wrote:

> Thanks, Neil.  Actually, I think I have one more question...  (Well,
> one question with many faces.)
> 
> Is the stripe cache write-back, or write-through?  If the latter, when
> does it get flushed to disk, exactly?
> 

Write-through.

However I am tempted to make it write-back as filesystems now submit 'flush'
requests which can be used to flush the cache.  I wonder if you would be able
to notice any performance difference...

NeilBrown

Re: Two questions about stripe_cache_size

[Roberto Spadim]

could we select it with sysfs?

2011/2/22 NeilBrown <neilb@suse.de>:
> On Tue, 22 Feb 2011 14:13:14 -0800 "Patrick J. LoPresti" <lopresti@gmail.com>
> wrote:
>
>> Thanks, Neil.  Actually, I think I have one more question...  (Well,
>> one question with many faces.)
>>
>> Is the stripe cache write-back, or write-through?  If the latter, when
>> does it get flushed to disk, exactly?
>>
>
> Write-through.
>
> However I am tempted to make it write-back as filesystems now submit 'flush'
> requests which can be used to flush the cache.  I wonder if you would be able
> to notice any performance difference...
>
> NeilBrown
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>



-- 
Roberto Spadim
Spadim Technology / SPAEmpresarial
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Two questions about stripe_cache_size

[Patrick J. LoPresti]

On Tue, Feb 22, 2011 at 2:24 PM, NeilBrown <neilb@suse.de> wrote:
>
>
> Write-through.
>
> However I am tempted to make it write-back as filesystems now submit 'flush'
> requests which can be used to flush the cache.  I wonder if you would be able
> to notice any performance difference...

I imagine it would depend on your file system and what you are doing with it.

In my application, I am typically generating multi-gigabyte files,
with linear writes, at a reasonably fast pace.  So write-back might
help.

On the other hand, "flush" alone might not be sufficient for high
performance.  I do not know much about Linux file system internals,
but it seems to me that there are two separate but related concepts.
One is ordering of operations (to ensure file system consistency).
The other is flushing (for data integrity; i.e. fsync()).  If the file
system actually has to flush to ensure consistency -- rather than just
tagging its writes with ordering constraints -- I think you might lose
a lot of your performance to all of that flushing.

But since I really do not know what the kernel provides internally in
this regard I will stop rambling now.

 - Pat
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html