R
Richard Shulman
Guest
In general there would be somewhat random performance loss if the two 4k OS blocks were not contiguous when we go to retrieve the block. This would require an OS disk reposition to complete the acquisition of the single database block. If there weren't many of these then the loss to performance would be miniscule. If there are many of these then you might be sacrificing about 5ms every reposition. Example of bad scenario. 1) Reading 100000 db blocks. 2) 80% of these blocks do not reside on contiguous OS blocks. 3) If normal read speeds are 5ms to read the two contiguous blocks and we have to do 20,000 of those connected block reads that is 100 seconds overall. (20,000 * 5 / 1000) 4) Now we have to read 80,000 of those non-contiguous blocks so it is 5 ms + 5 ms * 80,000 / 1000 = 800 seconds. The second 5ms is for the reposition to get to the second block. If we didn't have the reposition time it would have been 400 seconds. So there could be some benefit if the file system block size matches the database block size if the underlying blocks we use for the database are not contiguous.
Continue reading...
Continue reading...