On Tue, Apr 19, 2005 at 02:23:44PM -0400, Griggs, Donald wrote: > I'm not sure how your network is set up, but if you're using 100mbps > ethernet, the full- table-scan queries are only 4-5 times slower over the > net. Wouldn't that be the same order of magnitude as the ratio of link > speed to ATA local disk bus speed?
Possibly. Although as I noted, our "real" application is showing a speed differential of about 50x. Much more than can be attributed to simple transfer rates. But it sounds like the answer is the crappy NFS locking speed. > As for the inserts, if you're achieving over 36,000 of these per second, are > you perhaps in a CPU-bound state rather than a disk-bound one? I think, as I indicated to Jay earlier, that the reason the insert speed doesn't degrade is that I'm performing a transaction around all the inserts. Thus I expect that I only pay the NFS locking price once rather than multiple times when I'm doing the query. > Given that there's no sqlite server software on the network host, scanning > full tables 50 or 60 times a second sounds awfully fast to me --- though I'm > an admitted newcomer to databases and may have some faulty logic here. > > I wonder if placing sqlite on the host, and adding something like SQRELAY > would be worth the time to setup? If your queries are much faster, you > might consider keeping it, or using a more conventional database like > postgres or mysql. Nope. that isn't going to be an option for me. Preserving the "zero configuration" and "single file" environment is of utmost importance in our application. Sounds like I'm pretty much stuck. -- William Hachfeld
