Eva 4400 DR performance very slow

[leew]

Hi.

We have experienced a failure yesterday evening whereby our Solaris hosts and one Windows Exchange server lost connectivity to Vdisks presented from our EVA4400. We are currently running f/w 09522000 and are planning to upgrade on Sunday to 09534000. HP state the problem is due to the firmware, however I also believe it could be down to CA. We replicate Synchronously as our WAN link RTT was <2ms average. However recently noticed that RTT can peak at over 40-60ms. Is this to blame? Would setting the DR groups to Async get around this issue?

[Troy Klement]

We have two EVA 4400s with embedded 8GB switches and CA for each. We have just started testing CA and have successfully used both synchronous and asynchronous. However, we have dark fiber connecting the switch fabrics directly using LW SFPs. So the problem could be in the ISL connectivity.

There is a known issue with 09522000 and CA. We are running 09534000 on both of our EVAs and have FC switch firmware v6.3.2.

[leew]

Since upgrading to the latest f/w we have yet to see the issue re-occur, fingers crossed :-)

[Athar Iqbal]

I have exact same issue with the same configuration but EVA8100 not 4400. Looking for solutions

[Heather O'Neil]

We're seeing the very same problem. Initially we were replicating using (2) EVA4400s, but now one of the EVA4400s was replaced with an EVA8400. We have a T3 (43 Mb/s) link and an average RTT of 35-40ms. We're doing async replication only. Both arrays are currently running XCS code 09534000.

There definitely seems to be an issue with the EVA4400s. Is anyone successfully replicating between EVA4400s across an FCIP tunnel?

Thanks,
-Heather

[Anthony Silverman]

We also have two sites each with an EVA4400, CA license, Brocade switch and mpx110. Replication is set up between the two sites in asynchronous mode.
The bandwith between the 2 sites is 100MB.
We have Winodws servers, VM machines and VMS servers connected to the EVA4400 replicating to our DRP site.

Our setup works fine under most conditions.

However, if something causes a large amount of data to be written at our production site in a short space of time, eg. creating a new VM machine writes 36GB of data, then the data replication group log starts to fill up. While the EVA's are emptying out the log the entire bandwidth of 100MB between our two sites is used. While the log is emptying response time on all servers connected to the production EVA drops to such a bad level that all our users start to complain and batch jobs run 50 time slower. (one hour instead of one minute)

HP has not provided an adequate solution. I would like QoS on EVA to give top priority to servers and lower priority to replication.

Any ideas anyone?

[Heather O'Neil]

Anthony,

Would you mind sharing what your zoning looks like? I am curious to see your zoning configuration and compare it with ours. I'll be doing some tweaking in two weeks to hope to resolve the response time issue. What we're running into is that sometimes the command view server response time comes to a crawl, especially when replication is occurring. We keep on getting "excessive ping" errors.

Thanks,
-Heather

[Anthony Silverman]

Heather,

We also had some 'hanging' problems with CV but we have upgraded to CV 9.2 and it is much much better.

I have attached our zone config. (with a few changes to keep the CSO happy)

It may look a bit messy becaue we are in the middle of an upgrade. Currently we have 6 (3x2) brocade switches. I hope that by the weekend we will be down to 2. After the changes, if the zone config looks a lot neater, I will attach the new zone config.