Recovery time

[Benoît]

Hi,

Does somebody have any idea/experience about recovery time for Oracle database?
Depending on volume (MB or GB) of Archive Logs for exemple ...

Thanks

[Patrick Wallek]

It is going to depend on a whole lot of factors that you give no data about.

How are you recovering? From tape? From disk? Are you using RMAN? How was the backup done?
How big is the DB? How many archive logs do you have to reapply? How big are the archive logs?

At this point no answer can really be given other than it could possibly take only a few minutes or it could a few hours or days.

[Benoît]

Let say : recover database using backup controlfile until cancel
With about 40 Archive Logs files on disk drive for about 20 GB total / biggests 1 GB

Database is about 100 GB (Oracle 9i)

I just need an average rate or samples.

Thanks

[Rick Garland]

To get a base value, what type of hardware - tape drives, are you using?

Waht is the read time for those tape drives?

[Benoît]

Disk drives are 36 GB 15K FC on VA7410 (full)

[Benoît]

Server is rp8400 with 2GB FC cards

[TwoProc · Accepted solution]

Last time we benchmarked this a couple of years ago - on a PA8700 (750 Mhz) system using an XP512 with 32Gig of cache in the box we could recover at an average rate of 1 500 Meg archive file / min. But we've found that the rate varies (as was pointed out) VERY much with the processor speed and the speed of the disk system.

[Benoît]

Thanks for this answer.

Any other sample would be welcome.

[Devender Khatana]

Hi,

We also have oracle running in archive mode. And we require to apply logs on two servers at a time. One SD16K having 12 CPUs & 16GB RAM takes 3 minutes to apply a 200 MB log with XP512 having 6GB cache.

Another L2000 with 3CPU's & 6GB RAM with same XP takes 5 minutes to apply the same log.

HTH,
Devender

[TwoProc]

I just noticed more info about machine class, and card speeds on your posting, so I can give more similar data: Database size was about 120G back then, 4 2GB FC cards were used to connect to SAN, 8-2G connections from SAN to XP512. Disks were spread across all available controllers and evaluated Raid 5 for worst case - used 27-4packs of R5 disk across 4ACP's. 10K rpm drives. SGA was approx 5G with about 3.8G for cache_buffers, if memory serves me well. Test was conducted with all other servers connected to the SAN down. 16 processors @750Mhz with 16 db-writers(io slaves actually). Effect of increasing/decreasing number of db-writers for recovery was untested(stuck with planned production settings). Eval result (from this and other tests) was that archive log disks require R0/1 at peak loads despite write speed gains from large cache size. Assumed that Raid configuration of archive log area wasn't involved in speed tests of database recovery as archive drives only experience read loads during recovery.

[TwoProc]

Correction, I just spoke with another co-worker(another dba) from our dept, and she says that the data that I remember is largely correct, except that back then, we used 400Meg files - not 500. So, it was an average of 400Mg/Min - not 500Mg/Min.

[Yogeeraj_1]

hi,

on this setup:
RP5430 x 2
Oracle 8i + RMAN
100Mbps link

and
Database size 30GB
Archive log file size 25MB

we obtained the following results:
Full RMAN recovery = 2 hours
Archived Log recovery (x1) = 15mins

(NB. on the system where we were recovering, the datafiles are on internal disks whereas the source backup files are on a SAN and recovery done across the 100Mbps network connection to the source server)

hope this helps!
regards
yogeeraj

[Benoît]

Thanks guys!
This should be useful we more details to compare on my config.

The thing is that we were supposed to make that test on test rp8400 that just crashed.
So I try to get information on average speeds for that king of recovery besause I can't make that test right now.

To give you more information, my server as 6 CPU 750MHz and 6 GB of memory.
I think we have just one db-writer.

Maybe it would be just more simple to measure I/Os my production can produce...
How can I do that ?

[Yogeeraj_1]

hi,

maybe glance and measureware can help you here!

regards
yogeeraj

[Volker Borowski]

Hi,

I asked nearly the same question here:
http://forums2.itrc.hp.com/service/forums/questionanswer.do?threadId=220331

Meanwhile I think it would be good to have a seperate configuration for recovery:

db_writers = num_cpus - 4
DB-buffers = 80% RAM

do parallel recovery (4)

May be do some tests to balance the CPUs between the parallel recovery processes and the db_writers.

Good luck
Volker