Re: Fast Inserts and Hardware Questions

Orion Henry writes:

> The indexes will be
> int8 (primary key)
> int4 (group number)
> timestamp (creation date)
> int4 (customer id)

Since one query can only use one index per table, you should only need the
customer_id index, given the queries you listed.

> * Is there an OS that is best suited for postgres. All things being
> equal I would like to run this on Linux.
> * Is there an architecture best suited for this. Should I shell out the
> extra cash for a 64 bit box over a 32bit one. Quad Xeon or Quad Alpha?
> Quad Alpha or Quad UltraSparc?
> * Since most of what I am doing is inserts I will assume that the disk
> will be my bottleneck over anything else. Is this correct?

Probably. The file system is also going to be relevant. Linux' ext2fs is
not the best possible choice here. The CPU is probably going to be the
least of your problems.

> * Will the 7.1 WAL save me, when it comes to insert times?

To be expected.

> * I read something about disabling fsync() to speed up inserts. How
> much will this speed things up? I would consider it as I COULD rebuild
> lost data from my logs in the event of a system crash and one night a
> year of panic on my part is worth saving $100,000 in drive arrays.

Disabling fsync can be considered obsolete with 7.1 WAL.

--
Peter Eisentraut peter_e(at)gmx(dot)net http://yi.org/peter-e/