Slow performance (or unexpected rates) with 10Gb and SSD storage arrays

[mjp]

Write back won't be enabled on the controllers - they're setup for FastPath (LSI)... That requires:
Write Policy: Write Through
IO Policy: Direct IO
Read Policy: No Read Ahead
Disk Cache Policy: Disabled

But, even with Caching enabled, it's the same... (though we lose performance since FastPath becomes disabled)

[gharding]

We are getting a similar issue, and have noticed when doing an iometer test we can get good write performance without replication but with replication we lose about 50% performance. We test with windows product and the replication almost max's out the 10GB used for replication.

We have tried different raid controllers from LSI, and Adaptec all with SSD cache.

We have tried the DRDB tuning and jumbo frames but no improvement

Any suggestions would be appreciated as we are wanting to put this into production but cant until we have this resolved

[dagar]

I too would be interested in a fix for this. I have noticed some speed decreases when introducing replication, but have not gotten as far as really benchmarking it.
Thanks
Derek

[To-M]

Send in the log file to support and provide as much details as you can.

[mjp]

I've finally had the chance - I created tickets for 2 different clusters that have speed and performance issues in similar setups.

[dagar]

I had went off of this for my DRBD tuning options:
http://kb.open-e.com/Tuning-recommen...1GbE_1603.html

Just wondering if there are better tweaks for direct connected 10GBe?
Thanks
Derek

[To-M]

The 1GbE article is the same for 10GbE, try to disable scatter-gatter and TCP segmentation offload since in the Console screen go to CTRL ALT W then select Tuning options then select TCP offload options to turn those 2 options off. I heard that this might help for some Intel 10GbE NICs is to turn off the DCA Direct Cache Access, you can turn this off at the Console in CTRL ALT T then select Modify driver options and select the 10GbE NIC.