Truly funny. So I start banging on this problem with truss. I notice that at some point sshd forks off a shell to run 'locale -a'. I notice that is taking several seconds longer on the slow sshd host than the other one. So I try this:
time locale -a > /dev/null It takes a fraction of a second on host A and several seconds on host B. I mutter a few choice expletives and start wondering if there is some kind of performance b0rkage on host B. I start groveling through zfs properties for rpool on each host. Ready for the punchline? I was only able to upgrade the RAM assigned to the OI VM on the main hypervisor a few weeks ago, and until then, RAM (and therefore ARC) was tight, so (drumroll) I had set primary and secondary cache to 'off' on rpool. I set both properties back to 'all' and run the locale test again. No change (and no surprise). Run it again (and the locale files should now be in ARC), and... Runs like greased lightning. Fire up an ssh session to host B. No more slow logins. Boy, that was fun :) _______________________________________________ OpenIndiana-discuss mailing list [email protected] http://openindiana.org/mailman/listinfo/openindiana-discuss
