Hello Dmitry, > Tuesday, September 11, 2012, 8:11:20 PM, you wrote: > >>>>> In sum, according to trace: While the 75 page buffers restore took >>>>> 601030ms, the 100000 page buffers restore took 375253ms. > > LS> Interesting. > LS> Could you try one more test, with cache size = 40,000. > > I did restore test for some production database, and found > no difference for db cache from 1k to 16k.
Did you use the service API (-service in case of gbak) when restoring to bypass the TCP stack? > http://www.ibase.ru/devinfo/rs-win-se-cache.gif > > -i - index creation turned of. > So, I do not see difference in it neither for tables, nor for indices. I don't see any significant difference with 1K - 16K for both tables and indices either, but I do see much better performance (~ 50% improvement in execution time, thanks Kjell *g*) for index creation with a largish page cache (> 40000) but only for *large* indices and it seems, for FK indices only, but I have to check. > TS> With a higher page cache, one can see in Task Manager a steady increase > TS> of RAM usage while restoring records, which *might* get re-used during > TS> index creation, because it doesn't need to pull data pages from the disk > TS> again when creating the index. > > I think here OS cache is more sufficient, especially for databases > that bigger than RAM. Here one thought came to me, that was inspired > by Vlad Horsun - to restore all indices per table, table by table, > not that way like gbak restores indices in current versions (all > foreign keys goes at the end). Right. I think FK indices creation benefits from a larger page cache, cause they are created at the end of the restore process. gbak restore performance is bound to the performance of a single core/CPU, thus running the index creation process in parallel, sounds encouraging. This is even getting more important (worse) with a faster I/O system (SSD ...). I'm not talking about utilizing several cores for a single index creation statement though (which would be cool, but is an area for Oracle and the PARALLEL clause). Dmitry, do you have any chance to try IB XE3? Thanks, Thomas ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
