One of the weaknesses of Bricolage seems to be the STFP connection.
For those who haven't used it, it's very slow on large media files.
We have a a 44MB video that takes a minute and a half to publish to
the production server with bric_queued. 1-2MB image files also take
10 seconds or more. Once you get 200 people publishing, well, that
causes a backlog.
We've installed Math::BigIng::GMP, and that is primarily for reducing
login time - we're fine on that. The actual file transfer is what
seems to be slow.
At any rate, we're looking right at Net::SFTP as the potential
culprit. Anyone have any thoughts or experiences with it not living
up to expectations? Is there something we might be doing wrong? The
Bricolage server and the production server are both hooked into the
gigabit backbone. Bandwidth is not a problem, and that's been
verified with other tests.
We're thinking about rewriting Bric::Util::Trans::SFTP.pm to link up
to Net::SFTP::Foreign, which uses the machine's ssh client that it
find in the path, and seems to get much better transfer times. The
main issue I can see with that is that Net::SFTP handles usernames and
passwords better than Net::SFTP::Foreign, the latter requiring Expect
code to set up the password exchange.
Any thoughts on this? Net::SFTP::Foreign seems to be pretty new, and
I doubt it would have been considered when SFTP functionality was added.
-Matt
For those who haven't used it, it's very slow on large media files.
We have a a 44MB video that takes a minute and a half to publish to
the production server with bric_queued. 1-2MB image files also take
10 seconds or more. Once you get 200 people publishing, well, that
causes a backlog.
We've installed Math::BigIng::GMP, and that is primarily for reducing
login time - we're fine on that. The actual file transfer is what
seems to be slow.
At any rate, we're looking right at Net::SFTP as the potential
culprit. Anyone have any thoughts or experiences with it not living
up to expectations? Is there something we might be doing wrong? The
Bricolage server and the production server are both hooked into the
gigabit backbone. Bandwidth is not a problem, and that's been
verified with other tests.
We're thinking about rewriting Bric::Util::Trans::SFTP.pm to link up
to Net::SFTP::Foreign, which uses the machine's ssh client that it
find in the path, and seems to get much better transfer times. The
main issue I can see with that is that Net::SFTP handles usernames and
passwords better than Net::SFTP::Foreign, the latter requiring Expect
code to set up the password exchange.
Any thoughts on this? Net::SFTP::Foreign seems to be pretty new, and
I doubt it would have been considered when SFTP functionality was added.
-Matt