RE: Antw: filesystem considerations

[Julien Huon <Julien.Huon@aduneo.com>]

Hello Ulrich,

Interesting. Did you try JFS ?

Thanks you.

Regards

-----Message d'origine-----
De : openldap-technical [mailto:openldap-technical-bounces@openldap.org] De la part de Ulrich Windl
Envoyé : vendredi 24 octobre 2014 08:17
À : Julien Huon; openldap-technical@openldap.org
Objet : Antw: filesystem considerations

>>> Julien Huon <Julien.Huon@aduneo.com> schrieb am 23.10.2014 um 18:06 
>>> in
Nachricht <571e5e67216d40a0b3943a660e73955c@VATICAN.aduneo.com>:
> Hello,
> 
> 
> I read a long time ago theses very interesting benchmarks : 
> http://symas.com/mdb/microbench/july/.
> 
> It seams that ext2 is the fastest filesystem for asynchronous writes.
> 
> Do you think an ext4 filesystem without journalisation but with 
> extents could be faster ?
> 
> A LMBD database is a big flat file isn't it ? Extents are very 
> interesting with that kind of files, aren't they ?

Hi!

I din't try ext4, because it's unsupported in SLES11, but I did a comparison of other filesystems for SLES11 SP2. Ext3 definitely isn't the worst choice; ext2 may be even faster. However xfs also performs quite well on large files. Usually the larger the blocks you are writing, the higher the throughput. And you must distinguish whether the buffer cache can buffer the asynchronous write or not. It also depends on the stroage technology you are using. I did my tests on a 4Gb FC-SAN with a RAID5 distributed on many disks (>20). ext3 also had peak read performance.

But for LMBD you would have to test mmap() operations specifically. Numbers may vary from traditional I/O operations.

Regards,
Ulrich

> 
> Did someone try it ?
> 
> 
> Regards