I have an 80gb IODRIVE sitting in an HP ML350 G5 server running Windows 2003 64bit os.
I will shortly be installing the Progress Openedge database software and migrating our production database from a virtual machine back to physical (on the IODRIVE).
Now, the recommended block size for the IODRIVE is 4096. The recommended block size (performance wise) for Openedge is 8192 and that the underlying disk block size be the same. Does anyone know WHY Fusio-IO recommend the 4096 block-size? Is there an underlying TECHNICAL reason or does it just represent a good compromise for "general" applications. Ideally, I want the best performance from OpenEdge AND IODRIVE.
On the other hand, does anyone know WHY Openedge recommend the 8192 block size? Is it to keep disk reads to a minimum (because traditional disk reads are slow?), if so I can stick with 4096 for the database blocks AND the disk blocks.
Our application is 90% reads and 10% writes (or something in that order) so READ performance is much more important that write performance (hence interest in Flash storage). So, propeller heads, which is better: 4096 or 8192?
You're dead-on, Barnsey... we pre-format with 4kb block sizes simply because we feel it's the best possible compromise for the widest range of uses OOB.
Thanks Terrence. I will post a performance summary once our software suppliers pull their fingers out!
Hi Guys,
well, after a long wait we finally got around to migrating our databases to the new Fusion. Wow. This has completely alterered the way we do business - tasks/reports that used to take 30-40 minutes now either take either a few seconds or a minute or so. We no longer have processes slowing each other down. We don't have customers hanging on the phone whilst we await their details to pop up and everything operates the way things should. This has boosted productivity and efficiency within the organization as well as reducing stress levels among the staff!
To get here though we've had to de-virtualize the database server which was a bit of a pain and so, to some degree we lose some of the advantages of VMWARE. However, the raw performance makes up for all that. Previously we were running VM's on an IBM FC SAN (using 15K SAS drives, and X3650 servers - all brand new in november 2007).
Database Block sizes. We've retained the 8K block sizes for the NTFS db partition - so we have databases of 4K blocks and 8K blocks on the same disk and there is absolutely no performance penalty that we can see. It could be argued that we hadn't squeezed enough juice out of the existing SAN configuration, it could be argued that we hadn't optimized our db storage and it could also be argued that we hadn't employed a Progress/Openedge consultant to get the very most out of the entire setup - but these things are all time consuming and expensive. far cheaper to throw hardware at it - for 2k GBP who's complaining?