EVA 8000 Managing Controller

[Mark Poeschl_2]

We have an EVA 8000 with a large number of vdisks presented to some UNIX hosts. All have "preferred path" set to controller A or B Failover/Failback with about half the vdisks preferring controller A and half on B. Strangely, a significant majority of the vdisks report "Managing Controller" as controller B. Why are some of the disks that prefer controller A managed by controller B? Is there some kind of intelligent load balancing going on between the controllers? EVAPerf shows that controller A's CPU is much more likely to be the busier one during our peak times.

[Rob Leadbeater]

Hi Mark,

Can you be more specific on the versions of UNIX that you're running ?

Cheers,

Rob

[Mark Poeschl_2]

Tru64 5.1B-5

[IBaltay]

Hi,
the issue could be the following:
a) after the swaping the managing controller role to the proxy controller and failback to the original managing controller, the host continues to write/read to the proxy controller and after a certain period of time it forces the role of the managing controller back to the proxy. Thats why the controllers need not be well balanced. Check the multipathing solution for the possible reasons to this behaviour... What UNIX are you using?
If HP-UX, then what version and what version of the multipathing?

[IBaltay]

Hi,
this could give some info...
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01552954/c01552954.pdf

[IBaltay]

Hi,
or maybe here for more docs:
http://h30097.www3.hp.com/docs/pub_page/V51B_DOCS/v51b_doclist.htm

[Uwe Zessin]

That version should properly detect optimized/non-optimized paths, but maybe the handling was disabled:

http://h30097.www3.hp.com/unix/erp/c01627023.html

[IBaltay · Accepted solution]

Hi,

New cam Attribute Controls Path Usage
http://h30097.www3.hp.com/docs/patch/51B/bl27/HTML/ch01s29.html?jumpid=reg_R1002_USEN

[Mark Poeschl_2]

Our cam_ccfg_aa_enable parameter is set to the default value of 1 which, as I understand from previous posts and the 'man' page should mean we're only using preferred paths. That doesn't sem to be the case.

Everybody's consensus seems to be that this is an issue with the UNIX servers rather than the EVA itself. Do I have that right?

[Uwe Zessin]

Mark, you have checked that the operating system type on the Host object is set correctly for Tru64?

[Mark Poeschl_2]

Yes. I've made that mistake in the past and as I recall things just didn't work at all if you had that wrong. (That may have changed with later versions of VCS/XCS.) In any case, the hosts are indeed set to OS type = Tru64

[IBaltay]

Hi,
cam_ccfg_aa_enable=1 gives us reads only via the managing controller
cam_ccfg_aa_enable=0 reads go also via the proxy controller

the problem could be that Tru64 even if there is SET PREFERRED PATH TO CONTROLLER A or B, FAILOVER/FAILBACK, is not able to correctly detect the failback of the controller e.g. after the repair and it continues to read via the proxy controller. If more then 60% of the READ IO (minimum 5000 IO per hour) within the next hour, the EVA controllers swap the managing contorller role back to the proxy controller. Thus after the time you can have e.g. 80% utilisation of the controller A and 20% utilisation of the controller B.

[sgparker]

Mark, was there any one resolution to your issue in July?
I am seeing the same issues in a mixed OS enviroment EVA8000, over 95% of vdisks are managed by one controller.
Thanks for any info.