Alexandre Benson Smith wrote:

>>>
>> AFAIK even on FB 1.0 this 'bug' was never really confirmed. And I have
>> heard reports of no problem at all on fb 1.5 (can anyone comfirm?!)
>>
>
> I won't call it a bug, but a limitation on the algorithm that cause
> slowness if a big (> 10k) pages on cache is used. AFAIK 1.0 and 1.5 has
> the same algorithm to handle the page cache. FB 2.0 has changed it.
>

There was a bug - the "b-tree" that managed write ordering of pages
from caches was degenerate - effectively a linear scan - and got more
significant as the cache grew until at about 10,000 pages increasing
the cache made the database measurably slower. I think Nickolay may
have fixed 1.5, Jim fixed Vulcan.

Regards,


Ann