I am using syslog-ng 1.4.7 and libol 0.2.20 and am attempting to do some tests to make sure that I can get the level of performance that I need. I am having the worst trouble though trying to isolate whether a problem is occuring in either syslog-ng or within the networking code of the machines that I am using. I've managed to hack things in the network code for some of the machines to get better performance for some of the tests, but we have dropped down the requirements now. I am sending along some test code that should compile on either red hat 6.2, red hat 7.0, and hopefully others. (Haven't tried) The problem I am running into is that I lose anywhere between 30 and 800 messages. As the message number goes up and number of clients, I lose more messages. Some results I have gotten, with 100 threads connecting to the server, I can send 60,000 messages in about 30 seconds without a problem (200 bytes), but when I increase the number to 300 threads, and I have it sending for 30 seconds for a total of 36,000 messages, I start to lose messages. Notice in the first case I was sending more messages and had it sleeping for less time between instances, in the second case it is sleeping longer and sending less messages per second per thread. I have tried it with multiple different machines and types of varying speeds, and it happens on all of them. For the latest run, I have for teh sending machine an AMD athlon 1Ghz w 256M RAM, with an ATA 100 Hard drive running on ATA66 running red hat 7.0. For the server, I have a PII 300Mhz with 180M RAM, a ATA33 hard drive running without acceleration. The server is running red hat 6.2. I have tried the numbers with the sender and receiver switched, but I lose more messages that way. If you look in the code at logconf.h, you can change the #defines. The logtest.c just creates the threads. log_thread.c has the send_messages function that simulates a syslog-ng client. If anyone can provide any help, it would be greatly appreciated, since I have to be able to show this working before I can get the full sign-off to use it for our the applications we are building. Matthew M. Copeland Software Engineer matthew.copeland@honeywell.com