MSA Storage
cancel
Showing results for 
Search instead for 
Did you mean: 

eva8000 can't keep up with management requests, help!

 
Scott Scriven
Occasional Visitor

eva8000 can't keep up with management requests, help!

I've been doing some SAN management with storage on a few EVA units, and keep running into a problem -- the EVA cannot keep up with management commands.

Basically, I want to batch create/destroy hundreds of vdisks and LUN presentations on a daily basis. The EVAs do not seem designed for my usage pattern, and they just fall over under the load.

So, anyone have ideas for how to avoid breaking it, and/or communication venues to report problems I've found?
10 REPLIES 10
Uwe Zessin
Honored Contributor

Re: eva8000 can't keep up with management requests, help!

Scott,
is there a specific reason why you cannot re-use existing virtual disks?
.

Re: eva8000 can't keep up with management requests, help!

Not that I can fathom why anyone would want to create/destroy hundreds of vdisks on a daily basis, but, how about converting to Containers and back?

Note: While I am an HPE Employee, all of my comments (whether noted or not), are my own and are not any official representation of the company

Accept or Kudo

Uwe Zessin
Honored Contributor

Re: eva8000 can't keep up with management requests, help!

How do you convert a container into an (empty) virtual disk? At least in CV-EVA/SSSU V6.0.2 connected to an EVA with XCS V6.000 I can't find anything that suggests this is possible.
.
Scott Scriven
Occasional Visitor

Re: eva8000 can't keep up with management requests, help!

The idea of re-using vdisks might be feasible. I'll have to think on it... It doesn't look promising at first glance, but it may be workable. It would eliminate the most expensive management operations, at least.

However, it seems to introduce quite a bit of complexity, and may only help a little. The bulk of the load is coming from I/O, and that won't change. It's common for I/O speeds to get as slow as 300KiB/s over a 4Gb/s link when a few people are hammering it at the same time.

Another option I'm considering is slowing down my SAN switches, untrunking them or downgrading port speeds... the idea being more load on the switches and less on the EVAs.
Mark...
Honored Contributor

Re: eva8000 can't keep up with management requests, help!

Hi,

A container is an empty disk structure until it is used in a snapclone or mirrorclone. The EVA has "mapped" out the Psegs/Rstores ready for use.

Using containers would save a lot of overhead on the EVA as suggested above.

Using something like RSM to script/schedule this may be of use?

No need to delete recreate Vdisks, just umount and turn into a container - just re-use the containers.

Mark...
if you have nothing useful to say, say nothing...
Chris Rosan
Valued Contributor

Re: eva8000 can't keep up with management requests, help!

Scott,

Why do you need to continually destroy the vdisk's? Do you need to erase the data or do you just need an "empty" disk/lun/volume?

What O/S are your hosts?

I can't fathom an app that you need to do such a quantity of those sort of operations on a daily basis.

There may be a way to do what you're doing at os level.

Re: eva8000 can't keep up with management requests, help!

Uwe, you are right.
Yes, you can not convert a container directly back to a vdisk. They are to expedite the creation of snapshots, snapclones and mirrorclones. Which I suspect would play a key part in whatever Scott is trying to do.

Scott, what ARE you attempting to accomplish?

Note: While I am an HPE Employee, all of my comments (whether noted or not), are my own and are not any official representation of the company

Accept or Kudo

Scott Scriven
Occasional Visitor

Re: eva8000 can't keep up with management requests, help!

> Scott, what ARE you attempting to accomplish?

I figured this would come up. No sane person would do this.

The answer is... testing servers. The EVA only serves to stress-test fiber channel, and LUNs are only needed for the duration of a test.

Unfortunately, I'm trying to break servers, not EVAs. :)

Re: eva8000 can't keep up with management requests, help!

LMAO!
OK, I've seen some rather interesting tests in the past. The best was years ago. A friend was working on the OpenVMS system crash/cluster recovery code. He had one system configured to boot, join the cluster and then crash itself, all in a tight loop.

Anyway: You say "...LUNs are only needed for the duration of a test...." I've always been rather fussy about this when setting up a new EVA. LUNs only exist once the virtual disk is PRESENTED to a host. No presentation, no LUN. Which brings up the questions:
Can you reuse a virtual disk, and just present the vdisk for the duration of the test and unpresent after?
What would care if the virtual disk had already been used?

Note: While I am an HPE Employee, all of my comments (whether noted or not), are my own and are not any official representation of the company

Accept or Kudo

Scott Scriven
Occasional Visitor

Re: eva8000 can't keep up with management requests, help!

I've now got a list of 9 things to try and reusing vdisks is one of those items. It's one of the more promising options, actually. It just requires a fair amount of effort (code changes and testing) before I can find out whether it helps. So, I'll wait for results from some easier options first.

However, I can tell already that I'm in for a bumpy ride... in the past hour I had an eva8000 completely ignore all management requests (even easy ones like adding a host or listing objects) for about 20 minutes. I think it simply isn't designed for the amount of load it's getting.