We have some devices mapped to FA ports 4a:0 and 13a:0 on the DMX-3 and allocated to the AIX host. We have another set of devices from FA ports 3c:1 and 14c:1 from the same DMX-3 and allocated to the same AIX host. We are running into some performance issues on the FA port utilization we decided to map the devices which are on 3c:1 and 14c:1 to 4a:0 and 13a:0. and then present to the host. If everthing is ok then cleanup the FA ports 3c:1 and 14c:1 . Is it doable when the AIX box in in production. Do we need any downtime. Please clarify.
You do not say what version of AIX you are running or what type of multipathing you are using, or if your host is clustered. To perform the steps you need I would recommend that the host be offline and iin a position to power down/reboot.
Your host will still have device links to the 'old' path to the devices, so your ODM could become confused if not cleaned up properly when the same device is presented from another FA
You will have to re-zone the switches, and activate this zoneset
You could run into reservation issues at the FA level if your host had reserved the device already. (The reservation is done at an FA level)
Hope this helps.
Thanks for your suggestions..
The AIX version is 5.3 and we are using EMC powerpath...Its not a cluster..I didnt get about the zoning thing.. why the zoning has to be done.?
From DMX -3, 76 devices were mapped to FA port 4a:0 and 13a:0 and then masked to the AIX host and 184 devcies were mapped to FA ports 3cb:14cb and then masked to the AIX host. So, total 260 devices were alloacated to the AIX host from the same DMX frame. So, they have perfromace issues on the nightly oracle jobs which are taking longer time to run. We see some utilization more on 3cb and 14cb so what i did is mapped all the 184 devices on 3cb and 14cb to 4a:0 and 13a:0. and then masked to the host when the host was up and running. Now in the powerpath it shows
260 devices under FA 4a:0 and 13a:0
184 devices under FA 3c:1 and 14c:1
Is this the right approach. Even after this changes the host are still having performance issues.
If I understand you correctly, now 184 devices have 4 paths to the host, via 4 FA's and 76 devices have 2 paths to the host via 2 FA's.
You do not say how many HBA's etc you have.
From a powerpath load/balancing perspective, yes this will allow PP to balance a load accross all four paths, but as to wether this will resolve your issue. It really depends where the contention lies. And determining if your bottleneck is host/SAN or Symm based.
Performance issues are hard to track down as they are usually found after a process of elimination.
I would recommend:
1) Run a grabfile on your host Grabfile tool available from ftp.emc.com/pub/emcgrab/Unix )
2) Run it through HEAT (available on powerlink http://powerlink.emc.com) and ensure all suggested parameters are met (HBA driver and ODM especially)
3) Check error reports for SCSI errors.
Then open a call with the Service Team, they can health check the Symm, SAN and host again,
If performance issue still occurring and all health checks clear then the local performance team will be called in to analyse the environment.
Hope this helps,