On Fri, 2010-01-22 at 09:41 +0000, John Haxby wrote: > 2010/1/22 Robert G. (Doc) Savage <[email protected]> > > I've been trying F12's vinagre as a means of remoting a > RHEL5.4 server's display to a central console. A full screen > display can be remoted with vinagre, but the VNC connection is > quite slow and pausey even over a Gigabit Ethernet. This > slowdown is not just on the remote F12 viewing console. The > connection is driving one of four cores on the RHEL5.4 server > to almost 100%, dramatically dragging performance down on its > local console. Here's a snapshot of "top" several minutes > *after* the vinagre connection is dropped: > > top - 23:57:46 up 6 days, 8:02, 5 users, load average: > 2.81, 2.55, 2.47 > Tasks: 253 total, 3 running, 250 sleeping, 0 stopped, 0 > zombie > Cpu(s): 13.2%us, 2.2%sy, 0.0%ni, 59.7%id, 0.0%wa, 0.1%hi, > 24.8%si, 0.0%st > Mem: 8175136k total, 8135640k used, 39496k free, > 191424k buffers > Swap: 2032212k total, 216k used, 2031996k free, > 6328248k cached > > PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ > COMMAND > 9 root 39 19 0 0 0 R 99.7 0.0 213:12.23 > ksoftirqd/2 > 4521 root 17 0 276m 58m 20m R 52.2 0.7 923:43.59 > Xorg > 4812 doc 15 0 316m 22m 15m S 3.0 0.3 285:21.38 > vino-server > [snip] > > Can someone suggest a safe way to bring the load average back > to its "pre-vinagre" value of ~0.05. The "kill -9" command is > ineffective on ksoftirqd. > > > I don't know what is causing the problem with vinagre, but it's worth > checking the gvnc web site for update and you might want to try the > sample gvncviewer.py to see if you get the same problem. The problem > might actually be with vino (to which vino-server belongs) that > behaves badly when its connection gets dropped. > > Fortunately, you can't kill ksoftirqd as it's a kernel thread (if you > could kill it your machine would hang terminally within a very short > time). Try killing vino-server and that should stop it doing > whatever it is doing to make Xorg occupy as much CPU time as ksoftirqd > will let it. If it doesn't, then kill Xorg as well. >
John, Thank you very much. Killing vino-server allowed ksoftirqd/2 to cool. After about 10 minutes the load average dipped below 0.10 and things now look rather normal. I'll check out that sample gvcnviewer.py. I'm a bit unsure just who to file a Bugzilla report with, RHEL or F12. I wonder if both groups might work together on this since F12 is supposedly going to be the foundation for RHEL5. --Doc Savage Fairview Heights, IL
_______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list
