@TRiG I appreciate the advice; I can see how editing the question instead of appending info is a better approach. The information is scattered through the comments and it's a bit difficult to keep track of. Right now I can't edit it, though, and besides, I moved to a new VPS provider which fixed the problem (confirming it was in fact the provider) so the question is moot now.

I switched VPS providers. The free command now shows what I expect it to, and the tunnel does not fail. I used the exact same configuration script and nothing changed on the tunneling machine. So there was DEFINITELY something wrong. Whether the weird memory usage was directly related or a symptom of a larger problem with the host, I guess we'll never know. The only major differences are the virtualization software and the kernel version used by each.

Thanks for the explanation. I'm assuming at this point that it has to do with OpenVZ since as I've said, I've run a lot of Linux machines before without this problem. (I actually created a CentOS VM instance with 2GB of RAM and configured it with the same configuration script, then ran the tunnels against that VM instead of the VPS... the free output showed that it behaved VERY differently than the VPS I'm asking about in the question, and instead behaved like the VPS has done for the months before I noticed the change). I'll just continue to monitor the logs and see what happens...

Good advice to increase log verbosity, I just did that for sshd and I'll see what happens next time the tunnel fails. Is there a way to do the same thing and somehow log messages related to memory/cache errors?

In that case, why would the SSH tunnel be repeatedly failing with no log messages, with that behavior appearing at the same time as the high cache usage? That's not normal, and that's what I'm trying to debug. If the high cache usage isn't related, then it's an incredibly strange coincidence. Given a lack of any information about the SSH failure, I'm hoping finding the source of the cache usage might give me a clue as to the source of the SSH errors. That's why I asked the question.

The critical thing to remember is that it never filled the cache like this before a couple weeks ago, and nothing has changed in the meantime. It's not about the cache being filled - it's about why it's suddenly doing it now and not before, and why at the same time the SSH tunnel is repeatedly failing with no messages or warnings (I've checked /var/log/secure and there's nothing that indicates a failure - it's a silent failure that requires a reboot before any more SSH tunnels can be created).

But 2GB of cache is used less than 30 seconds after a reboot, where there is far less than 2GB on disk and nothing being sent through the tunnel. The VPS control panel reports cache as part of its memory usage, and prior to a couple weeks ago always reported low usage. That's my question - why would it suddenly begin using large amounts of cache where it didn't before, and where there simply isn't enough data to fill the cache anyway?