Squid http accel as a front-end buffer.

Is it appropriate to use Squid in http accel mode as a front-end buffer to
buffer between a heavyweight server (such as Apache with mod_perl)
serving uncachable content and potentially very slow remote users (such as
users behind slow modems or wireless modems)?

I've attempted to measure whether it's appropriate. I wrote a small TCP
redirector with the ability to pause between packets, simulating a slow
connection. For small documents (50k or so), it seems to work fine, the
server finishes serving the document in a couple seconds, while it takes
the client a couple minutes to finish downloading. But for larger
documents (250k or so), it doesn't work at all, the server takes a couple
minutes, while the client takes more like eight minutes. More precisely,
it works after a fashion, but the "raw" server serves the data up in 10
seconds or so, so it doesn't work nearly as well as it should.

I suspect that why it doesn't work so well for large documents is that
when a document isn't cachable, it only devotes a limited amount of buffer
space to it.

In any case, is this an appropriate application of Squid? If so, are
there any parameters which would apply to tuning this problem?

Thanks,
scott
Received on Fri Apr 23 1999 - 17:02:35 MDT