MirrorDisk performance hit during initial synchronization

[Rudy Williams]

Hello--

Can anyone tell me what sort of performance hit (a percentage) is exptected on an HP-9000 server (11i) during initial mirror synchronization? In other words, what sort of a performance hit will I take on the system when I issue the command "lvextend -m 1 /dev/vg00/lvol5"?

[Pete Randall]

That's one of those "that depends" questions. My reaction is that the performance hit should be minimal. There is a lot of I/O that's going to be going on but it sounds, given your use of /dev/vg00/lvol5, like you're mirroring your root VG, where there should not normally be all that much I/O normally. It should be fine.


Pete

[James R. Ferguson]

Hi Rudy:

Synchronizing a mirror does take some time and will reduce performance slightly. Your milage may vary. Synchronization is performed logical volume extent by extent, so the elapsed time is proportional to the size of the logical volume.

Regards!

...JRF...

[A. Clay Stephenson · Accepted solution]

It's noticeable but things don't slow to a crawl or even nearly so. I would put the performance impact somewhere between 0% and 100% with an extremely high confidence level (1.0). With a confidence level of 0.7, I would put the impact at about 20% + or - 15%.

Whenever I have to replace failed drives, the users never really notice a performance degradation and the duration is on the order of a few tens of minutes.

[Jaime Bolanos Rojas.]

Rudy, looks that performance when creating a mirror or replacing a bad drive is always a concerned.
Some logical volumes are small some are really big, but even thought there is not a specific command to check on performance, you will hardly notice it.

What I would do, just for curiosity, would be to run the sar -d command before you do the mirror and then during it, so you can see how the values change.

Regards,

Jaime.

[Rudy Williams]

Folks--

I completely understand the "that depends" answer on the performance hit. But are there any rough/general numbers that we can expect to see? 10%? 25%? 50%?

We don't want to mirror the lvols during production hours if it is going to greatly affect system performance.

[Rudy Williams]

Wow, hot topic and fast replies. Two posts came in before I could follow-up!

Thanks for the information all (esp. Clay).

[Steven E. Protter]

This is perfectly normal. When you extend its a very heavy I/O operation.

SEP

[Tim Nelson]

Keep in mind the risk that exists of the primary failing while trying to decide when to sync the mirror.

Do it now.

[Dave Wherry]

If your system is so stressed now that mirroring an lvol will give it a noticeable (to the users) performance hit, it's time for an upgrade.