Darren Reed wrote: > I've always considered sockets to support asynchronous I/O through > non-blocking I/O and using occassionally with signals. > > How is what you're proposing any different from what already exists?
Scaling. Existing facilities don't scale past about 10^4 sockets per process (and you're into extended poll for that; plain poll and select don't really go beyond 10^3). Nor do they do well with high throughput randomly arriving on many endpoints, though I don't have figures for that. The ICSC work was aiming for the region of 10^5 sockets. Cheers, Jeremy Harris _______________________________________________ networking-discuss mailing list networking-discuss@opensolaris.org
