Resolving when EVA components fail

[Hanh_3]

Hi,
We have EVA5000 with single-path configuration, without Secure Path. Some Hosts take LUNs controlled by Controller A and the other hosts take LUNs controlled by Controller B.
My first question is: If Controller A fail, all LUNs it controls with fail-over to Controller B. (I don't know whether it support in single-path enviroment, without Secure Path)
Then we just change the Zoning Configuration so that Servers effected can see their LUN again. Can it possible ?
Second question: Controller A is fixed, all the LUNs will fail-back to Controller A ???


We have 2 Disk Enclosure, 13 FC disks anf 15 FATA disks, configed as 2 disk groups, Sparing level is None, all vdisks are Vraid5.
Third question: What is maximum number of failed disks acceptable?
As far as I know, there is 1 to 11 disk per RSS, so each of our disk group will have 2 RSSs. So the maximum number of failed disk for each disk group is 2 ?

[Bostjan Kosi]

Hi,

In windows environment your zoning should be set only to see single path to the VDISK. Of course in case of the controller failure you can change zoning to allow server to connect to one of the ports on the other controller. Your app will fail, since you can not do this fast enough. In FC switch create two configs and then just enable the correct one. Of course after replacing the controller you just enable again the correct config on the FC switch. Be shure to select no preference for vdisks fasilover policie. So the EVA will show the disk over the path which is available. You'll probably also loose disk conectivity for a sec or two when failling back, so make shure no apps are acessing the disks at the time.

3. Yes you are right in teory, just make shure the correct two disks fail in your system :-)

BR
B.

BR
B.

[SAKET_5]

Hanh,

You need a failover/failback driver for the operation that you expect - should a controller fail, LUNs failover to the other controller - when the original controller comes back - the LUNs perform a failback.

You would need to use some sort of driver/software to have this capability. For Tru64 UNIX and VMS, the OS has the capability of path failover natively - with the EVAs you can also set this up from the controller end.

A popular software to perform such functions is Secure Path with dual Hba, dual fabrics, etc.

For single path, if one of your controller fails, you lose access to all your LUNs and your servers would crash failing to write to those suddenly disappeared LUNs!! Yes, you could configure zoning and then get back into business - but this is messy!

Failback is again a driver/software feature.

Depending on how the RSSs are setup, vraid 5 should sustain the loss of one disk in each RSSs. All your vraid 0s will be lost tho.

Hope, this helps.

[SAKET_5]

Hanh,

just for completeness of information, you could look into migrating to a dual redundant setup with the use of some free software that is available these days to perform path failover/failback from the host end. Yes, cost of the HBAs and the fabrics would still be there - but you would migrate to a much cleaner and resilient environment!

For Linux - using Qlogic FCAs - you can use their Qlogic failover driver - FREE

For Windows - you could use the HP supplied MPIO driver - does path failover - FREE

For Solaris - you could look into MPxIO - Free

Tru 64 and VMS, as suggested just use the EVA functionality of failback and failover - dont need NO software!

Hope this helps and consider assigning points to the responses to show your appreciation.

Regards,
Saket.

[Hanh_3]

Thanks Bostjan

Yes, our servers are all Windows.
"Be shure to select no preference for vdisks fasilover policie. So the EVA will show the disk over the path which is available. "

Do you mean that when a server is zoned to see 1 port of a controller, and server's vdisk is set to "No preference", then EVA will automatically choose that Controller to control the vdisk ?? And it vdisk will fail-over and fail-back as you say ??
I'm never working with single-path environment, and as I know, after create vdisk, I cannot change "Path preference" property, so really want to make sure before we create the vdisks and set properties for them.

[Hanh_3]

Thanks SAKET
So in our situation, Windows server, single path, without SecurePath or MPIO, that means no failover/failback driver, so vdisks cannot perform failover/failback. We will loose all vdisk controlled by the failed controller and must recreate those vdisks ???

[Bostjan Kosi · Accepted solution]

Just for clarification:Since you do not have redundant paths the whole discussion is based on assumption that you'll loose curent conectivity to a presented disk from the OS point of view - meaning, apps will stop and you'll have to either rescan disks or reboot the server. You are not planing to purchuse another HBA (& FC switch) nor Secure Path. With Secure Path or as SAKET sugested MPIO driver even using just one HBA you would survive a controller failure without any interference.

EVA controllers look for a familiar WWN of the HBA. If EVA finds already defined WWN (Hosts - ports) It'll show the created and presented (to that WWN) disks no meter what the path will be. As long as there is visibility (logical - meaning you have correct zones to link together the port where HBA is connected and at least (in your case just one) one port on EVA system) the server will be able to access the disk.

So there is no fear on having to recreate vdisks or something like that. Just a path modification will be necesary and a possible disk rescan or server reboot required. This is the cost of saving money by not buying a redundant solution for a fully redundant storage system :-)) As much as I remember you can change the Preffered path/mode setting without no problem on a live and working vdisk. Just go to presentation tab an select the correct type and click save changes..

[Hanh_3]

Thanks Bostjan,
Your answer help me much.
As you suggest, we will set vdisk to "No preference" and just do load balancing between 2 Controllers through Zoning. (We know the cost of saving money, but this is our situation).

[Bostjan Kosi]

Good to hear. Please close the thread when possible.

BR
B.

[Chris Watson]

Saket,

I have been unable to locate the setup information to use mpio under Windows (for free). Have you got any good links?