topic FCIP Replication performance problems in MSA Storage

FCIP Replication performance problems

BR711326 — Thu, 19 Oct 2006 08:37:32 GMT

I'm looking for some light in this CA installation:

EVA3000 - VCS 3028
EVA4000 - XCS 5031
Two independent fiber links (dedicated to CA) with bulk electronic in it, that delivers 30-75 Mbits/sec. Length is less than 10 Km.
2 + 2 SR2122 with only FCIP protocol, dedicated to CA.

In the performance test phase we use Command View to manage CA activity. No SRM installed.

Ping letency for 2KBytes packets are 4 ms (round trip). FTP bandwith is over 8MB/sec average. We measure this using two hosts that use this line.

Using the performance estimator we should have arround 100-115 I/Os sec (block=16KB) and trip latency of 2ms and low line usage.

Now the reality ... very sad data:
- Full Copy of one single copy set in one DR group takes for ever ... 0,6 MB/sec. Using EVAperf drt we get latency of more than 1000 ms.

- Host I/O (iometer) of 16KB (full writes) gives arround 2 I/Os/sec

During the test there are no other significant load on the EVA3000 and NO load at all on the EVA4000. The line is dedicated to CA all the time.

The obvious answer is that the problem is the line speed. Said that, way the ping & ftp numbers are good and the CA are so bad ??
Have you had similar experience in other FCIP replications ??

Any help or experience sharing will be appreciated.

Re: FCIP Replication performance problems

eis1953 — Thu, 19 Oct 2006 16:57:18 GMT

How are you converting FC to IP on Brocade MPR's, if so make sure you have exchange trunking turned off.

Re: FCIP Replication performance problems

BR711326 — Fri, 20 Oct 2006 01:25:02 GMT

The FC - TCPIP gateway is the SR2122 from HP (actually is an old Cisco NS54xx).
We don't use the Brocade MPR.
The switches are SAN switch 2/16 & 2/16V.

Re: FCIP Replication performance problems

BR711326 — Fri, 20 Oct 2006 01:34:39 GMT

The solution:

Reading the SR2122 Getting Started Guide
P/N=AA-RTE1G-TE
page 62 ...
regarding the parameter "useBport"
you can see:

Use Fibre Channel (FC) B_port connectivity. This configuration option applies
to all FCIP connection types. By default, the useBport setting is enabled.
However, the setting is dependent on the Fibre Channel switches and switch
firmware in use:
â B-series
â useBport disabled (B-series FW 3.2.x or 4.4.x or later)
â useBport enabled (other B-series switch FW)
â C-series - useBport disabled (MDS FW v2.01b or later)
â M-series - useBport enabled

My switches versions are 5.1.x & 3.2.x so I have to disable useBport.

After that I get 250 I/O/sec doing iometer on a replicated disk and arround 4MB/sec bandwidth.

So, take care with this parameters !!

Re: FCIP Replication performance problems

BR711326 — Tue, 17 Apr 2007 16:44:38 GMT

Same installation, new problems:
From time to time (once or twice per week) we got communications problems between the SR's.
Link drop and after few second/minutes link come up again.
We have two lines, 4 SR2122-2 for FCIP.
In one of our lines, as soon as the line comes up, we can see automatic FLOGI from the SR and the CA activity resumes.
In the other line, when the link comes up, nothing happend.--> connection down for ever.

To fix that, I need to disable/enable the FC switch port where the SR is, in this way the SR logs in again to the fabric.

As far as I can see, all the elements in the two lines are similar. Way then this big difference reaction?
Where can I start debugging this ?

Thanks in advance.
antonio

Re: FCIP Replication performance problems

Shaun Walsh — Mon, 28 May 2007 19:31:01 GMT

Try this, It was just released

http://h18006.www1.hp.com/products/storageworks/fcipd_gateway/index.html?jumpid=reg_R1002_USEN

Shaun