Failover using multipathd on Linux 5.4

[Tim Barton]

Hi,

Since HP has told us we need to start using the native mulipathd and Qlogic HBA driver starting with Linux 5.3, I've been trying it on a test box running Oracle Enterprise Linux 5.4 64-bit, using two QLogic HBA's to an EVA 8100.

I've got multipathd working in theory, but it takes a long time to do a failover. I'll run an "ls -l" on a SAN-mouted LUN after disabling one of the two fiber ports that the server is connecting to, and it will take 1-2 minutes before the command works.

Is this expected behavior, or is there a way to make it shorter? I don't want to risk downtime if an Oracle database on one of these LUNs loses connectivity. I've played with some of the multipathd.conf settings without much luck.

Thanks.

[Michael Leu · Accepted solution]

Hi Tim

The Release Notes of HP's own Device Mapper Enablement Kit has a section with the recommended device parameters:
http://h20000.www2.hp.com/bizsupport/TechSupport/SoftwareDescription.jsp?prodTypeId=12169&prodSeriesId=3559651&swItem=co-77761-1&prodNameId=3559652&swEnvOID=4035&swLang=13&taskId=135&mode=4&idx=0#Device_par_val

Scroll down a bit to see the QLogic HBA parameters. I think these also influence the failover time. They recommend "ql2xmaxqdepth=16 qlport_down_retry=10 ql2xloginretrycount=30" with the native QLogic drivers.

[skt_skt]

what are the default values you have for those parms? Even with default it should not take this much time.

I doubt if you need to tune at qlogic level as the communication between Multipath+array settings should take care of this

[Tim Barton]

Have this in modprobe.conf:

options qla2xxx ql2xmaxqdepth=16 ql2xloginretrycount=30 qlport_down_retry=64


Tried taking the qlport_down_retry parameter, but that didn't help. I'll try changing it to 10

[skt_skt]

I have qlport_down_retry set to 30. Most of my servers has the hba/qlogic level fail over set to false and fail over is takan care by power path/array. But i have only a very few machines running with multipath.

[Tim Barton]

Looks like those recommended settings did the trick. Failover response time is now just a few seconds, which should be fine.
Thanks!

[skt_skt]

did u tune all the three parms Tim mentioned? can u confirm what is the current values u have?