With regard to Q34 about SSD. There are vendors that have onboard RAM cache incorporated into their SSD.
The first vendor of SSD we looked at didn’t have this feature and when we benchmarked it, we found it was
indeed slower than mechanical disk. Would be interesting to hear comments about using SSD from a vendor
like BitMicro which has onboard RAM cache. We are planning on benchmarking this.

So I wouldn’t offload my log files to SSD unless the SSD incorporated an onboard RAM cache. Otherwise, as
Peter and Heiki say,there would be no benefit.

Note: A lot of the newer boxes SUN makes have internal SAS drives. If you already have external RAID arrays
with BBU write cache, then I don’t think it makes sense to offload the log files to these internal drives. If you
do, then you would be sacrificing a lot of performance. If you don’t have BBU write cache in an external RAID
array, then it might make sense to offload the log files to the internal drives.

In our performance monitoring patches we now implemented per query working set estimation. It would not be hard to implement global working set estimation as well. Imagine table space is 1000000 pages we allocate bitmap which holds 2 bits per page telling us if page was written to or read during certain timespan, say 1 minute. Now we can pretty easily set bits in this bitmap and dump it to disk (it is just 250K/minute) and when use some tools to analyze how many distinct pages are being written/read per given time interval and so estimate how different query ratios will impact it. One can also estimate with fixed memory hash similary as we do for queries.