Has anyone switched to NPTL and noticed a speed difference? If so, what seems faster? Has anyone run across any packages that don't work well with NPTL? What about the practical difference between + and - nptlonly?
I recently updated a Mysql box from Mysql 3.2.x, 2.4 kernel, dual PIII, and 2GB RAM to Mysql 4.0.x, 2.6 kernel and nptl, dual P4 Xeons, and 4GB of RAM. With hardware upgrade it's been hard to tell exectly what improvements were caused by which upgrades.
The db's themselves are around 4-5 GB in total so the new server can fit a significant percentage more of the Mysql data into RAM. We moved from Mysql 3.2.54 to 4.0.24. The disk I/O is U320 on the new server compared to U160 on the old, but both with RAID cards and similar RAID setups.
The old server was running 400-500 mysql threads, a number of apache boxes make calls to it, and averaging 300 queries/sec. The load would peak at around 2 with the number of threads being the main influnce of load.
The new server is doing the same number of queries and has not been completing them any faster than the old server. However load remains a steady 0.10-0.20 whether there are 300 threads or 500. This appears to be the main benefit of the new threading. Better scalabilty and more efficient use of resources in a highly threaded application rather than any outright speed increase. That may translate into a speed increase if NPTL helps eliminate any resource contentions you may have had.
kashani -- gentoo-user@gentoo.org mailing list
