Re: Benchmarking Sybase and Oracle? (No flame wars please)

Gary Lawrence Murphy wrote:

> Are there any meaningful benchmarks for dataservers? In particular,
> I am looking at hitting a database of several thousand records
> several hundred times per second from several hundred remote locations
> and wonder if this is even possible.
>
> Also, I understand Oracle and Sybase both have provisions for
> running parallel dataservers and I wonder how much effect this could
> have on the above scenario (since we would also be adding in the
> replication overhead to the transactions per second)

Oracle Parallel Server is not the same as replication. The only overhead
on parallel servers is if there are collisions at the db block level between instances on different machines. The Distributed Lock Manager handles this but it has to send a message to the other instance, wait for it to write the updated block to disk and then wait for the reply. If you can break up your user transactions to avoid block contention, you can effectively increase your processing power almost linearly by using the parallel server, up to the point where you can't add more processors or you become I/O bound on the disks.

There is a huge difference between transactions depending on whether they are large or small, query or update. It also depends on whether you are accessing the same blocks a lot (high cache hit ratio) or randomly accessing different blocks (low cache hit ratio). Check with Oracle, Sybase and/or the hardware vendor(s) for benchmarks. They will typically use TPB or TPC type benchmarks, but there is such a huge difference between transactions (as pointed out above) you can't really draw a direct comparison between the benchmark and your application. The only way to know for sure is to test it yourself, or find another application with similar characteristics that you can extrapolate from.

-- 
---
Allen Kirby			AT&T ITS Production Services
akirby_at_att.com			Alpharetta, GA.