[Lustre-discuss] Write optimizations on OSS

[Arifa Nisar]

Hi,

 

I have been running some write only benchmarks on Lustre with following two
versions at two different machines. 

lustre: 1.6.6

build:
1.6.6-19691231180000-PRISTINE-.usr.src.linux-2.6.18-92.1.10.el5_lustre.1.6.6
smp-perfctr

Configuration: 1 OSS for 11 OST.

 

 

lustre: 1.6.5

build:  1.6.5-2.1.50HDB_PS13A-rsrel-REL_2_1_50-19575-2009-11-13-18:58

Configuration: 1 OSS for 2 OST

 

I have noticed that if contiguous data is written to the OSTs under one OSS
then some performance benefit is achieved. Is there any caching, aggregation
or other optimization implemented in OSS for write accesses? I understand
that there is read caching but I could not get much information about write
optimization on OSS.

 

Thanks,

Arifa.

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20100308/e5740193/attachment.htm>