Re: New IBM Nonsense

I'm not sure I would call a TPCC disclosure document with 163 or 206 pages a simple scenario, but I've had a quick look through the documents.

The important points to note are that:

    the log files are configured to 150 MB each

    4 checkpoint occurred during the test

    the checkpoints were triggered (apparently)     by explicit timeouts, rather than log switches.

I note that the figures you quote seem to be the amount of space configured for log files, but there is no explicit report of redo log generated - only an upper limit implied by the checkpoint count - and the document suggests that this were forced on a timeout, not at logswitch.

In fact, the larger number you quote appears in a block of numbers describing space allocation for 60 days running, whereas the small number appears in a section that comments on the need for allocating logs for 8 hours.

-- 
Regards

Jonathan Lewis

http://www.jlcomp.demon.co.uk/faq/ind_faq.html
The Co-operative Oracle Users' FAQ

http://www.jlcomp.demon.co.uk/seminar.html
Public Appearances - schedule updated Dec 23rd 2004






"Serge Rielau" <srielau_at_ca.ibm.com> wrote in message 
news:35fj7cF4ki6acU1_at_individual.net...


> Jonathan Lewis wrote:

>>

>> If you can describe a simple scenario where the

>> logging done on RAC signficantly exceeds the

>> logging done on a non-RAC installation - with

>> an indication of data change volume, cross-instance

>> calls, and quantities of log generated, I can probably

>> tell you what's causing the difference. (And version

>> of Oracle, of course).

>>

>>

>

> I'm looking at the 1M+ TpmC Linux RAC result using O10g vs the 1M TpmC 

> Result without RAC on HP - also O10g

> http://www.tpc.org/results/FDR/TPCC/HP%20Integrity%20rx5670%20Cluster%2064P_FDR.pdf

> vs.

> http://www.tpc.org/results/FDR/TPCC/hp_tpcc_sd_1mil_fdr.pdf

>

> Apparently the Linux RAC required 24500 GB (24.5TB!) for 8 hours of 

> logging.

> The HP SMP result only 2,481.03 (2.5TB). That's a factor of 10.

>

> Now, if my understanding of RAC is correctly logging goes up for hot 

> pages. So I could imagine that the counter for the ORDERID may be nasty, 

> but not that nasty. Also only 10% of the orders are remote and the cluster 

> is partitioned.

>

> Cheers

> Serge

>