copying informix rawdevs using dd with bs greater 2k

[Martin Christov]

hello folks,

i am trying to copy 1TB informix db under hpux-11.00 from one emc-box to another using the command
dd if=rawdev1 of=rawdev2 bs=2k
if i do it this way i need appr. 40 hours to complete the copy of the 360 rawdevs and can not complete my action during the weekend.
is it possible to increase the bs-parameter for dd without danger of informix-datacorruption?

regards:
martin

[A. Clay Stephenson · Accepted solution]

Hi martin, in this case the dd blksize is totally unrelated to the Informix blocksize. Crank that puppy up to about 8000k. The last partial block will be handled just fine. If dd can only read let say 234 (1k) blocks at EOF then it will write 234 (1K) blocks. Also,
(you done good on one thing) don't specify an ibs and an obs - that will cause double buffering - just use bs.

Regards, Clay

[Martin Christov]

Hello Clay,
thanks for your answer!
At Sunday I've used my downtime and copied the 1TB using 4 parallel dd with bs=1MB over 4 FC and emc-powerpath-software(load balancing and failover).
The last copy finished after 4 hours.
Will an increase of the dd bs to 2MB, 4MB or higher lead to a decrease of the copy-runtime?
Next weekend is the next copytime.
Regards:
Martin

[Stefan Schulz]

Hi Martin,

a higher blocksize reduces the overhead for the copyaction. So you will see a decreas in the time needed for this copy. At least as long as you don't hit another performancebarrier e.g. network speed, IO speed of the discs and so on.

Hope this helps. Stefan

[A. Clay Stephenson]

Hi Martin:
I don't think you will see any significant increases if your buffer size is bigger. You might see a small improvement if you make you
bs a multiple of 64k. If you are a c programmer you might consider either a multi-threaded application or cooperating processes so that process A is reading to a shared memory buffer while process B is writing from
another portion of the same shared memory segement. You would have a moving window so that when the data is written out to disk it is available again for a new segment to be read in. If you look at one of the HP-UX Porting Centers you will find a program called 'buffer' which does just what I describe. You could pipe dd to and from it and try it.

I don't know if what you are really doing but I assume you are aware that EMC makes software to keep two arrays in sync.

Hope some of this helps, Clay

[Martin Christov]

Hello Clay,

I am going to migrate my database from 4GB splits (residing on emcbox1) to 32GB metavolumes (residing on emcbox2).
I am not trying to write better srdf or timefinder script ;-))

Regards:
Martin