Its About Time?

Some of the recently reported Socket I/O problems with UNIX Squeak 3.8a1 VM may be related to why Squeak tends to perform so badly as a webserver and why there have been challenges in getting a robust TeaTime working in Croquet on platforms other than Windows. Without going into the technical details (some of which can be seen in David Shafer's emails posted on the Squeak VM List), a Socket I/O problem cascades in a way that causes bad behavior in UNIX Squeak-based servers when those servers are receiving large numbers of closely spaced connections. In certain cases this problem may be responsible for intermittent bouts of long latency in Squeak VMs - of up to 1.5 seconds before an incoming packet is detected by the Socket plugin. At the same time, the Squeak VM appears to handle UI input events at rates of 400 per second or more without a problem. All of this may be related to why it has been a challenge to pull together a robust TeaTime in Croquet and it now appears necessary to do some platform specific network layer tuning of the VM as we close in on a working v 1.0 release of Croquet. David P. Reed has recently suggested that this aspect of Squeak performance may have gone unnoticed since it doesn't manifest in large file transfers which have been the primary form of benchmarking to date.