[Gluster-users] Read performances low compared to write (second try)

[Vijay Bellur]

On Friday 10 July 2015 02:18 PM, Olivier JAVAUX wrote:
> Hello,
>
> I am evaluating GlusterFS and I have a question about read performance compared to write performance.
> Release is 3.6.3-1 on SLES 11 SP 3.
> I have 12 storage nodes, configured in Ditributed-Replicated with 2 replicas.
> Network is 10 Gb ethernet.
> Filesystem used is xfs.
>
> When I am doing a simple test with dd bs=1M count=102400 (100 GB file with blocksize=1M), I have good perfs when writing:
>    around 550 MB/s - 10 Gb network is the bottleneck (2 replicas)
> but when reading, performances are less good: 470 MB/s
> I read one replica and read performances on the node itself might be better (around 800 MB/s)
> I tried several parameters:
>    performance.read-ahead: on
>    performance.read-ahead-page-count: 16
>    performance.io-thread-count: 32
>    performance.cache-size: 1GB
> I followed also performances hints for Linux kernel.
>
> Has somebody hints to increase these read performances?
>

Can you disable gluster's read-ahead translator and try again?

Additionally o/p of gluster volume info will be useful.

Regards,
Vijay