This is a very interesting concept that forces the Kernel (needs to be done on a system that supports TCP stack mods) to be unfair in its allocation of resources, by spitting out the requests that have the shortest time remaining first — in the case of a Web server, this means small files are sent before large files.
This is interesting, and not something that I have seen move beyond the academic literature. Is anyone out there using SRPT in their production Web serving environment? How did you determine that the predicted improvements in Web performance were in fact realized?
It also appears that their are some issues with SRPT, especially as file size increases. But it it would be interesting to get feedback from the Web performance community on this.