APerf(A Performance Measure) is a C library and distributed with some tools for the performance measurement. They are for HTTP, SMTP or LDAP servers and others. APerf library has two purposes, to generate a workload based upon strict time management under the non-realtime operating system or without realtime scheduling, and to manage concurrent sockets(i.e. network I/Os) with best efficiency.
For example, what will you do when you want to know your server to be expected having high performance is good enough for your service? Almost all people want to give a high workload to the server, and want to observe what will happen, such as the response is light and fast. But how do you generate such high workload? And how do you measure a time in the resolution of millisecond or microsecond under a heavy workload? As anyone can guess, generating a high workload needs machine power. If you were rich and can address this problem with buying more other machines, you might not need APerf library and tools. APerf library and tools can generate a high workload even on a poor and cheap PC, with measuring the response time.
The provided tools with APerf library are just sample code to show how APerf library functions are called. And they will produce a good statistical report to advise you to know the server application or the operating system.
The purpose of APerf library is very similar to the ancient event loop mechanisms, such as an implementation in postfix(mta) typically. But some protocols or libraries don't suit this mechanism. For example, the LDAP library hides the protocol under its implementation. In this case, our event loop will not work correctly. So APerf library supports the workload with pthread. APerf creates the threads as specified rate. hyperthreading?
If you are interested in this project, feel free to send any comment, objection to the author, Junjiro Okajima. And anyone can join the project development mailing list, aperf-devml.