What means "Powerpath trespasses LUN when all active paths are exhausted"?

this applies to clariion only (since its active-passive design)
powerpath always sends i/o on the path which go to owner SP for a lun (all the paths going to owner sp are active paths). i/os are not sent on passive paths unless the host has specifically used secondary paths (which is not the correct way). this is done to minimise trespass of a lun. (remember that an sp has to take ownership of a lun before the i/o can be handled by it and this take long time comprared to requesting i/o from alternative active path in case one of the active path has failed.)

normally the host will have to be configured to use the virtual powerpath device (e.g /dev/emcpowerX). due to this, the os and application will always send i/o on virtual device and then its upto powerpath to decide which paths to be used.

answer to your last question. if there are only two paths, and active path has failed, there wont be any chance of i/o going though that specific path. if you have configured your host to use powerpath device, it will never happen. so here powerpath will trespass the lun to alternative sp and i/os will continue on those path.

PS: when powerpath says paths are exhausted, i assume it means that paths have failed.

Many Thanks Kiran.

The document I said that mentioned that sentence is "PowerPath Foundations".

If is difficult for me to understand that exhausted means dead, but well, I'm Portuguese . I could only understand that exhausted is a cause for paths to get dead, which is a little bit different, but the behavior would be, as Kiran said, paths to get dead:

But if that is the sentence, does anyone knows exactly when PowerPath considers the path exhausted (meaning, the limit in queued I/O's for considering it dead)?


---
Just to clear some minds, exhausted as I know means the following:
1. To wear out completely.
2. To drain of resources or properties; deplete.
3. To use up completely.
4. To treat completely; cover thoroughly.
5. To draw out the contents of; drain.
6. To let out or draw off.

powerpath does two things
1. multipathing and failover - this allows it to recognise multiple paths going to a lun and then trying to use them so that failure of one path does not affect data access
2. load balancing - trying to use more than one paths to a lun so that more performance is achieved

powerpath can sense a dead path in two conditions
- an i/o is sent on one path but i/o fails due to path failure or timeout
- a periodic background process keeps checking available paths and decide if they are dead/alive

a path does not go dead (or exhausted) when i/o queue up on it...when i/o fails or timesout through a path, its considered dead. if this is the case, powerpath retries the same i/o on alternate path(s). when i/o fails on all alternate path, powerpath decides that the lun has ultimately lost contact.

now in case of load balancing, powerpath uses different algorithms to use all available paths. this policy (roundrobin, claropt, symopt, request etc) can be set by powerpath commands. depending on the policy in use, powerpath may decide to use alternate paths when load on a path increases.

this all should be documented in powerpath guide...

Thanks Kiran.

So, according with what you've said, a LUN will not trespass when path's get exhausted.

I'll request EMC to make some changes in the PowerPath Foundations manual regarding that sentence.

Thanks.

(word "dead" removed)
Message was edited by:
Celso Capao

LUN will not tresspass untill all active paths are dead...once active paths are dead, powerpath will initiate trespass..

I think the problem with the wording is a translation thing...

The original statement "When all active paths are exhausted" does relate directly to one of your definitions (resources depleted) in that PowerPath will attempt to use any active paths to a specific LUN. PowerPath treats the available active paths (whether there is one active path or ten) as a "pool" of possibilities. Once all those possibilities have failed it considers that "pool" as exhausted. If there is one active path and it fails then all possible active paths are exhausted. In a common configuration there will always be two active paths, and PowerPath will not consider the active paths as exhausted until both have failed.

Does this make more sense? They aren't really talking about individual paths in this statement, but in you case with only one active and one passive path to each LUN, it just turns out to be single paths.

"normally the host will have to be configured to use the virtual powerpath device (e.g /dev/emcpowerX)."

If the host is HP-UX/Solaris with VxVM. it dosent matter if you use a native device or a psudo device.
check this emc 83285. Pls let me know if otherwise..

I think that Allen gave the best possible explanation. The concept of "pool" of paths that PP can use really gives a nice and clear picture of what's going on ..

ThX Allen!

Yeah it comes down to the alternative meaningof the word exhausted in English, such as in the statement "I've exhausted all my options" meaning you've run out of options. In this case it means it's run out of available active paths.