On Fri, 9 Sep 2011, liuzl wrote:
> This means we can't receiving data until all the READ packets sent.
> is that ok?
No, that's not OK and we can't have it so. The best way (to get high transfer
speeds) is to send as many READ packets as possible before we read anything,
as on a fast network we might get read responses back once have saturated the
outgoing "queue".
> Why not send window adjust packet in _libssh2_channel_write() .
I think we should move out the window_adjust calling from within
channel_read() to have it done more independently. Doing that would also
remove the unexpected side effect that channel_read() will actually send data
on the channel at times. If we did only this, sftp_read() would immediately
get in a better shape as then the receive_window call won't destroy the
outgoing packets like now.
-- / daniel.haxx.se _______________________________________________ libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-develReceived on 2011-09-09