| Author |
Message |
caker
Joined: 15 Apr 2003
Posts: 2392
Location: Galloway, NJ
|
| Posted: Wed Feb 18, 2004 10:29 am Post subject: Host11 Reboot - 10:25 AM ET 02/18/2004 |
|
|
Host11 went under heavy disk I/O, and is unresponsive to commands. I'm preparing to reboot it.
-Chris |
|
| Back to top |
|
Devboy
Joined: 16 Feb 2004
Posts: 8
|
| Posted: Wed Feb 18, 2004 10:30 am Post subject: |
|
|
| Thanks for the info |
|
| Back to top |
|
nirvanis
Joined: 03 Jan 2004
Posts: 11
Location: Barcelona
|
| Posted: Wed Feb 18, 2004 10:59 am Post subject: Unable to boot my linode |
|
|
| I'm unable to boot my linode (66.220.1.83). |
|
| Back to top |
|
Devboy
Joined: 16 Feb 2004
Posts: 8
|
| Posted: Wed Feb 18, 2004 11:02 am Post subject: |
|
|
| the hub is in reboot! |
|
| Back to top |
|
nirvanis
Joined: 03 Jan 2004
Posts: 11
Location: Barcelona
|
| Posted: Wed Feb 18, 2004 11:15 am Post subject: Coming up |
|
|
| It seems to be booting now. |
|
| Back to top |
|
caker
Joined: 15 Apr 2003
Posts: 2392
Location: Galloway, NJ
|
| Posted: Wed Feb 18, 2004 11:21 am Post subject: |
|
|
Should be all set.
This was due to the crappy I/O scheduler in 2.4 kernels, which essentially DoS the box if one (or more) processes (like a UML kernel) decides to consume all the disk bandwidth.
I've been testing 2.6 kernels, so I thought we'd give it a shot:
Code: [root@host11 vm]# uname -a
Linux host11.linode.com 2.6.3-1 #1 SMP Wed Feb 18 10:18:43 EST 2004 i686 i686 i386 GNU/Linux
-Chris |
|
| Back to top |
|
nirvanis
Joined: 03 Jan 2004
Posts: 11
Location: Barcelona
|
| Posted: Wed Feb 18, 2004 11:32 am Post subject: Down again? |
|
|
| Down again? |
|
| Back to top |
|
caker
Joined: 15 Apr 2003
Posts: 2392
Location: Galloway, NJ
|
| Posted: Wed Feb 18, 2004 11:38 am Post subject: |
|
|
That didn't go well. Box froze (probably bugged/panic'ed) after a few minutes... Rebooting back into 2.4, should only be a moment.
-Chris |
|
| Back to top |
|
bji
Joined: 27 Aug 2003
Posts: 182
|
| Posted: Wed Feb 18, 2004 8:13 pm Post subject: |
|
|
caker wrote: Should be all set.
This was due to the crappy I/O scheduler in 2.4 kernels, which essentially DoS the box if one (or more) processes (like a UML kernel) decides to consume all the disk bandwidth.
I've been testing 2.6 kernels, so I thought we'd give it a shot:
Code: [root@host11 vm]# uname -a
Linux host11.linode.com 2.6.3-1 #1 SMP Wed Feb 18 10:18:43 EST 2004 i686 i686 i386 GNU/Linux
-Chris
I think that I/O scheduling is the single most important performance factor for Linodes, and that it should be a very high priority to get this problem solved.
As it stands, Linodes can easily hog all of the I/O bandwidth and the result is that everyone suffers quite badly. host5 seems to be experiencing bad I/O load a couple of times day; this might be recent as I just started tracking it, but I have seen about half a dozen occurrances per day of the load on my Linode going up to 3, 4, 5, or even 12 (!!!) for a few minues at a time. My Linode is unloaded so this is most likely due to some other Linode hammering the disk and processes in my Linode having to wait a long time for small disk requests (either paging in memory or touching the filesystem).
If the 2.6 kernel really does have a way to more fairly distribute the I/O load (so that no one Linode can totally hog all of the I/O), then I am so all for it! |
|
| Back to top |
|
| |
