Re: RAC High End Scalability?

In article <5b945v88h9rg40to39sab6cauio7r855gj_at_4ax.com>, Michael says...
>
>
>ORACLE make lots of claims about the scalability of RAC that I would
>like to believe.
>
>Microsoft recently (late last year) published a paper disputing some
>of these claims.
>
>http://www.microsoft.com/sql/evaluation/compare/RealityBehind_RAC.pdf
>
>Now, I'm well aware that a Microsoft paper is going to be biased, but
>they do make a good point that nobody have published any RAC
>benchmarks showing RAC scalability using a cluster of high end
>servers. I'm interested in knowing if RAC can be made to scale using
>small clusters (4 or 8) of IBM p690's, HP Superdomes or high end Sun
>gear, where each server has 16 or more CPUs in it?
>
>If a single 64-CPU HP Superdome can manage a tpmC result of
>423,414
>(http://www.tpc.org/tpcc/results/tpcc_perf_results.asp?resulttype=noncluster)
>
>what could a cluster of 4 or 8 of them manage with RAC?
>
>Does anybody have practical experience of using RAC for scalability
>(not just failover) in high end UNIX servers such as these?
>
>Regards,
>Michael.
>

Yes indeed. I implemented RAC on 3 (yes 3) Sun E10K's for a customer recently. Incredibly high throughput requirements (16000 tps) and it all worked very well. Only problem we had was when someone reorganized the database for some reason and left off the automatic space management clauses from the tablespaces and they all went back to one freelist and one freelist group. Talk about an easy way to kill performance! :(

We also run our mail servers for the entire company on a 3 node HP cluster. I think they might be Superdomes, but I could be wrong on that. There are a number of other examples I could cite you as well, but if you want to continue contact me offline so we don't bore anyone who might not be interested.

HTH. Additions and corrections welcome.

Pete

SELECT standard_disclaimer, witty_remark FROM company_requirements; Received on Tue Feb 18 2003 - 10:45:55 CST