%_load manual.var%
ConferenceRoom's dynamic multithreaded architecture allows it to take full advantage of multiprocessor machines. This architecture also protects ConferenceRoom servers from being ambushed by the operating system. Other products that suffer from 'ambush' can become bursty or unresponsive for brief periods of time as load increases. This multithreaded architecture also makes it easier to customize ConferenceRoom for your needs. ConferenceRoom can launch customer-supplied code to make decisions without the server having to wait for the customer's code to complete its operation.
ConferenceRoom is written in platform-independent code that sits on top of a small operating system adaptation layer. This allows the server itself to be equally positioned on all operating systems without having to emulate any one operating system's functionality on another. This also allows low-level code, such as network I/O, to be fully optimized for each supported platform. On Windows NT, 2000, and XP, I/O completion ports are supported for optimized scheduling and maximum I/O performance as the Windows kernel reads and writes network data directly to and from the application's buffers without the usual faults and copies. On Solaris 7 and greater, the '/dev/poll' interface is supported for highly efficient socket discovery. On Linux, LinuxThreads is used without the usual cancellation wrappers that sap performance on many system calls.
ConferenceRoom servers continuously monitor system performance and resource availability. In the event of resource shortages, the servers can run off of a private 'emergency pool' and continue to provide normal service as they trim their resource consumption. Servers monitor the size of their list of pending tasks and defer expensive tasks during bursts of high load. This automatic tuning allows the same product to minimize its impact on other system tasks in non-dedicated implementations but still provide the utmost performance for larger implementations. Generally, no application tuning is needed, and operating system tuning is usually limited to raising resource limits.