Buffer sizes should not be hardcoded, but be configurable or dynamically adjustable depending on the connection.

I believe they are settable via SetBufferSegmentSize/SetBufferMaxSize. The constants are just the defaults when you don't say anything else.

Yes but we need to adjust them based on the response from the site and/or prefs?

Oh -- I don't think adjusting them after calling AsyncRead currently works. We would have to give the ability to change them on nsPipe. It might optimize things slightly, but the main reason for the buffer size parameter was to throttle the amount of data that would be delivered to the main thread in any given OnDataAvailable call, thereby allowing it to return to the event loop and drive the UI. Optimizing the transfer rate by minimizing allocations is probably the least of our worries right now.

Moving to M17 which is now considered part of beta2.

pulling in ruslan's necko bugs ->darin

Hmmm... seems like it didn't really go to darin.

mass move, v2. qa to me.

while Bradley is looking at performance...

change milestone to 0.9.8 (from triage meeting)

i don't think this is going to be a critical performance/api issue for mozilla1.0 -> future

the number of buffers used is an issue for the main thread, but we don't need dynamically sized ones.