Qlogic FC HBA capped on proxy

[lars@norstat.no]

it varies a lot, i was down to 3 minutes when i did only a single job but it can be as much as 12 minutes when running several jobs at a time. Sometimes this is due to waiting for resources because of yet another stupid limitation with the NFC connection buffer at the vmware side.

At present i'm running 13 jobs continuous and in the evening i stop those jobs. Run the backup, one big replication job of 70 vm's and lastly the vcenter server. i have found out that there is sometimes problems with running replication while the Vcenter server itself is replicating so it gets it's own job. Then i start all over again.

The reason for having multiple jobs is that even though you now have parallel processing inside one job, when the first vm is finished it still have to wait for the whole job to finish before it can start again.

I'm going to run it for some hours now to see if it stabilizes a little bit better.

[emachabert]

Be sure to be up to date with vsphere 5.5 (U1 since last week) since there were multiple bugs regarding network drivers and other things.

If I understand well, you have v7000 on one side and v5000 on the other ? Why areńt you using builtin replication tools to achieve the best RPO ?

I agree hotadd is worse for writes and generate heavy small IO load...but it was for testing purpose only.

[lars@norstat.no]

Up to date yes.

SAN replication is expensive, lacks generations which they say they have now, but not in the same way. Also if you set up realtime replication the fast san slows down to the speed of the slowest san. Write have to be aknowledge at both places.

[dellock6]

And probably you are going to replicate by LUN, correct? VM replication is more granular, if you need this kind of control...

Luca.

[lars@norstat.no]

That's right. Some server could be replicated once a year and others continuously. And replicating to a thin disk gives more space on target for more generations.

[emachabert]

Storage design depends on your RTO/RPOs and application requirements, there's not one magical solution (not yet)

To be honest, I never went under the 15 minutes replication window with Veeam (using 3Par array with storage snapshot integration, thus creating LUNs for each RPO class).

[lars@norstat.no]

i have about 5 minutes on average now on the 13 servers i do all the time. Guess i should be happy with that then ....