Resident Memory

Hello iFolks, I have a question about the new functionality offered in the newer versions of IDS, where tables can reside in memory. If you have really good cache rates can an engine still benefit from this feature? I suspect checkpoints would perform better, as in not necessary for the hard drive writes, but it would seem like a marginal advantage to overall performance. So what is the real benefit if your cache rates are already high? In addition, has anyone heard of segmenting tables between memory and disk with this feature? It would seem hypothetically possible. Any ideas or comments would be appreciated.

My understanding of the memory resident buffering is that the table goes into high-priority buffers. What this means is that they will be the LAST choice of buffers to be re-used. The table is not loaded into memory automatically, but only the specific pages needed, which then remain 'forever'. So, if you have some pages of the table that are heavily accessed and the rest not accessed at all, the table has been effectively 'segmented' between disk and memory. With lots of buffers in proportion to your database, it probably does not add a lot to performance, but if you have several small lookup tables that can be quickly scanned, it might save on their access. And, of course, you are in essence guaranteeing that they will be in the buffer cache, even if normal LRU policy would have dumped the pages. Doug William Marquis wrote in message <7q6kk7$6ic$1@news.xmission.com>... > >Hello iFolks, > >I have a question about the new functionality offered in the newer versions >of IDS, where tables can reside in memory. If you have really good cache >rates can an engine still benefit from this feature? I suspect checkpoints >would perform better, as in not necessary for the hard drive writes, but it >would seem like a marginal advantage to overall performance. So what is the >real benefit if your cache rates are already high? > >In addition, has anyone heard of segmenting tables between memory and disk >with this feature? It would seem hypothetically possible. > >Any ideas or comments would be appreciated.