Re: [evlog-dev] evlog and db2
James Keniston <[email protected]> Wed, 5 Mar 2003 16:17:57 -0800
| Newsgroups | gmane.linux.kernel.event-logging |
|---|---|
| Message-ID | <OFC778F074.9D3D50DA-ON88256CE0.007CC547-88256CE1.0001A501@us.ibm.com> |
Question #1: evlview --new uses posix_log_notify() because posix_log_read() never blocks. Neither does posix_log_seek(), which is pertinent if you want only selected records. So you have to explicitly sleep() or pause() while waiting for the next record. The non-blocking nature of posix_log_read() can be a pain, but it's been in the POSIX spec since God-knows-when. Question #2: As I mentioned in previous (private) mail, I'd expect the plug-in technique to be faster. I just ran a little benchmark. On my laptop, I could log about 3800 NODATA events per second without evlview --new running. This dropped to about 2340 events per second with evlview --new running. (evlview managed to catch all the events, but evlogd simply logged fewer of them.) I attribute most of the difference to the notification overhead; presumably, evlogd had to spend more time waiting for evlnotifyd. Jim |---------+--------------------------------------------> | | Mark Megerian/Rochester/IBM@IBMUS| | | Sent by: | | | [email protected]| | | ceforge.net | | | | | | | | | 03/05/03 02:18 PM | | | | |---------+--------------------------------------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| | | | To: [email protected] | | cc: | | Subject: [evlog-dev] evlog and db2 | | | >--------------------------------------------------------------------------------------------------------------------------------------------------| I have made good progress with getting the evlog data into a db2 table. In looking through the code in evlview for getting notified of new events, I was puzzled by something. I suspect this is a naive question, but: - The code uses posix_log_notify_add to ask to be notified about new events, but then it uses posix_log_read to process all of the events. To the uninitiated, that seems like a contradiction. Why ask to be notified, if you are essentially going to loop on reading the log for new events anyway? Is it because there is a level of indirection between the evlog event, and this posix log?? Also, I am interested in getting any opinions on the performance expectations of using this notify technique vs. the plug-in technique. Expecting to have a large volume of events over 1,024 nodes, so pathlength is important. thanks! Mark Megerian (507) 253-1682 [email protected] ------------------------------------------------------- This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger for complex code. Debugging C/C++ programs can leave you feeling lost and disoriented. TotalView can help you find your way. Available on major UNIX and Linux platforms. Try it free. www.etnus.com _______________________________________________ evlog-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/evlog-developers ------------------------------------------------------- This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger for complex code. Debugging C/C++ programs can leave you feeling lost and disoriented. TotalView can help you find your way. Available on major UNIX and Linux platforms. Try it free. www.etnus.com