cancel
Showing results for 
Search instead for 
Did you mean: 

vxassist -b queuing behaviour

Hello,

I'm going to migrate data using the well documented volume mirroring procedure, adding a target mirror plex then removing the source plex.
Since there is 300 volumes i will use a script to create and sync the mirrors. I use the following command :

 vxassist -b -g [dg] mirror [vol]  [TARGET DISK LIST]  layout=nostripe 

I have a text file with 300 of these commands, which i launch using

 nohup sh mirror.txt &  

I noticed that not all the lines are processed instantly : a bunch of something like 20 vxassit mirror commands are processed, and can be seen in vxtask list, but then the shell waits, as if one of these commands was not returning. When one of the running vxassist completes,  the next one from mirror.txt is processed and so on.

My questions are :

- How is this queuing controlled, ie how many vxassist mirror can run in parallel ?

- Is this the correct way to do this ? Is there a risk that one of the "not returning" vxassist mirror will eventualy time out thus "skipping" to the next one and leaving me with a missing mirror ?

Thanks if you can advice or point me to relevant documentation !

 

Dave

1 Solution

Accepted Solutions
Accepted Solution!

I ran a trivial test

I ran a trivial test (5.0MP3/Sol10), and attempted to mirror 100 volumes (pretty small in size) in a similar fasion.

By default vxtask was only showing ~6 active tasks.

I then repeated but this time I slowed the resync operations, by inserting the -o slow in the vxassist command. This time I could many more actve tasks (40+).

On theory is that the active tasks are consuming vxconfigd resources, which are also needed to start up the new vxassist tasks.

cheers

tony

View solution in original post

5 Replies

Hi There are some throttling

Hi

There are some throttling mechanisms related to recovery operations (vxrecover) as too many can impact performance, though do not recall anything that would apply to your manually triggered operations.

Could tasks be finishing before new ones are starting up ? Do you have a snapshot of the vxtask list that can be shared

 

Also, what SF version and OS is being used ?

cheers

tony

Solaris 10 with SF 5. I'll

Solaris 10 with SF 5.

I'll grab a vxtask list when i get the chance, but it's kind of straightforward, with a bunch of vxassist mirrror running. In the meantime, the shell script that launches the vxassist is waiting for one of these to complete. The only possible reason i can see is that at a given time, a vxassist -b mirror is not returning to prompt, effectively waiting untill another one ends before getting processed.

As my colleagues say to

As my colleagues say to create a mirror is a procedure that requires a lot of cpu usage is why ifyou want to make a script that validates better ending to a hope that time and which start to do the other. if you give me the list of the server can I help you with the scripts

I would assume, how many

I would assume, how many vxassist mirrors can run at the same time would depend on number of vxconfigd threads (to be precide vxiod) threads .... but that would be something which is usually default (unless something is manually changed)

 

Not completely sure, but I would assume, number of vxassist processed would be number of threads it used (which won't be 300 to my knowledge), once queue get free then another is processed..

 

G

Accepted Solution!

I ran a trivial test

I ran a trivial test (5.0MP3/Sol10), and attempted to mirror 100 volumes (pretty small in size) in a similar fasion.

By default vxtask was only showing ~6 active tasks.

I then repeated but this time I slowed the resync operations, by inserting the -o slow in the vxassist command. This time I could many more actve tasks (40+).

On theory is that the active tasks are consuming vxconfigd resources, which are also needed to start up the new vxassist tasks.

cheers

tony

View solution in original post