On Wed, 24 Jan 2007, Mononen Jussi wrote: > One code based bottleneck has been identified long time ago, it was in the > packet.c and in the function packet_read which read only 16 bytes at a time. > Afaik this has been correct.
I have two entirely rewritten functions for the lower level packet read and write, so now we can tweak the buffer sizes to better match what is typically used in the protocol. Currently I use a 4096 bytes buffer for the reading. > As Sara mentioned in one of her earlier mails to this list other SSH > implementations (eg. OpenSSH sftp) use scattershot techniques (asynchronous > transfer) where several packets are sent immediately and responses (and > their ordering) is taken care of later when they arrive. This way the > bandwidth is utlized much more efficiently as it can be used to send data > while the remote end is processing previous messages. > > How to achieve this with an API like libssh2 is a problem to be solved. But > the current synchronous transfer is appr. 50% slower than OpenSSH sftp. I don't understand how the API is a limiting factor here. The application certainly will want the data to arrive in order no matter how the API is designed or how the internals work, and collecting data out-of-order can be done while still offering the existing API. Or perhaps I'm missing something... ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ libssh2-devel mailing list libssh2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libssh2-devel
