Hi,

I am doing some work around big datasets. My current table is over 1.3 billion rows (yet only 13GB) and working very nicely The server is single user with complex queries.

I would like to put this onto a specialised server. Any advice for a special bit of kit to build?  I was thinking of the following:

- A fast multi-core processor (any view on AMD versus Intel?) but don’t need multi-processor?
- 64-bit linux, probably CentOS
- Lots of memory, at least 32GB
- SAS controller for SAS 15K hard drives. Tables still sub-20GB so separate fast disks rather than the need for lots of storage

Any views?
Cheers
Paul

For a single user server, getting faster clock speed processors with less cores may be faster than getting a larger number of cores. Also, 32GB seems like overkill for 20 GB of data, but you may be able to experiment with putting the data directory on a ram disk for a read only copy.

CentOS 5.3 is a solid operating system and many companies are starting to use that as an alternative to Red Hat Enterprise Linux.

A faster drive won’t help much since we avoid I/O as much as possible but in my experience, read/writes when you do have to do them seem to happen about twice as fast on a 15K SAS drive vs. a 7,200 rpm SATA drive. For the ultimate in i/o speed, consider the new solid state disks, maybe put 3 of them together in a 2+1 RAID 5 configuration for optimal performance.