[lustre-discuss] POC Performance

Riccardo Veraldi Riccardo.Veraldi at cnaf.infn.it
Mon Apr 8 14:19:51 PDT 2019


it was not clear to me you were using 16K block from your email it
seemed like you were writing 16k files.
16k block size is really tiny. The default ZFS block size is 128K but if
you want to get performance out of Lustre/ZFS for big files you should
use 1M block size.
Anyway there are a bunch of other optimizations parameters to play with
Lustre (number of rpcs in flight and so on...)  and the  kernel module
parameters for zfs, ko2iblnd, ksocklnd, ptlrpc. You can gain even higher
performance setting the proper parameters.


On 4/8/19 1:09 PM, Bill Carlson wrote:
> Note the tests were run on a 20 cpu, 1TB (not GB) ram, also serving as
> the only OSS for the filesystem.
>
> I've noted a key item, sysbench tests were with 16k blocks. Changing
> to 1MB was much much better, only about 10% slower than raw disk.
>
> Time to move to the next phase. :)
>




More information about the lustre-discuss mailing list