Problem as follows:
1) Plug in an external USB drive.
2) Mount it anywhere. Doesn't matter how.
3) Copy a few GB of data to the drive from a non-USB disk.
4) Watch the load average "climb" to 5.x, sometimes 10.x or more. Why? This on an otherwise unloaded system. Doesn't matter how many cores, how much RAM, 32/64 bit, etc.
Why should copying some files to a USB drive cause load averages to climb so high? (and network monitors to freak out?)
On Apr 20, 2012 2:42 AM, "Lists" lists@benjamindsmith.com wrote:
Problem as follows:
Plug in an external USB drive.
Mount it anywhere. Doesn't matter how.
Copy a few GB of data to the drive from a non-USB disk.
Watch the load average "climb" to 5.x, sometimes 10.x or more. Why?
This on an otherwise unloaded system. Doesn't matter how many cores, how much RAM, 32/64 bit, etc.
Why should copying some files to a USB drive cause load averages to climb so high? (and network monitors to freak out?)
It's just a number. Is the system any slower?
Linux adds I/O wait time to the load average calculation.
-- Giovanni
On 04/20/2012 05:24 AM, Giovanni Tirloni wrote:
On Apr 20, 2012 2:42 AM, "Lists"lists@benjamindsmith.com wrote:
Problem as follows:
Plug in an external USB drive.
Mount it anywhere. Doesn't matter how.
Copy a few GB of data to the drive from a non-USB disk.
Watch the load average "climb" to 5.x, sometimes 10.x or more. Why?
This on an otherwise unloaded system. Doesn't matter how many cores, how much RAM, 32/64 bit, etc.
Why should copying some files to a USB drive cause load averages to climb so high? (and network monitors to freak out?)
It's just a number. Is the system any slower?
Linux adds I/O wait time to the load average calculation.
Problem isn't so much actual "speed" but causing network monitors to freak out due to "high" load average when performing backups. I can make exceptions for servers doing backups, but then I don't get notifications when the load is legitimately high. I can make exceptions only during backup times, but that increases complexity.
Seems silly that load average would climb to 2.x or more copying some files on an otherwise lightly loaded server.
On Thu, Apr 26, 2012 at 10:18 PM, Lists lists@benjamindsmith.com wrote:
On 04/20/2012 05:24 AM, Giovanni Tirloni wrote:
On Apr 20, 2012 2:42 AM, "Lists"lists@benjamindsmith.com wrote:
Problem as follows:
Plug in an external USB drive.
Mount it anywhere. Doesn't matter how.
Copy a few GB of data to the drive from a non-USB disk.
Watch the load average "climb" to 5.x, sometimes 10.x or more. Why?
This on an otherwise unloaded system. Doesn't matter how many cores, how much RAM, 32/64 bit, etc.
Why should copying some files to a USB drive cause load averages to climb so high? (and network monitors to freak out?)
It's just a number. Is the system any slower?
Linux adds I/O wait time to the load average calculation.
Problem isn't so much actual "speed" but causing network monitors to freak out due to "high" load average when performing backups. I can make exceptions for servers doing backups, but then I don't get notifications when the load is legitimately high. I can make exceptions only during backup times, but that increases complexity.
Seems silly that load average would climb to 2.x or more copying some files on an otherwise lightly loaded server.
You might be better off monitoring CPU usage instead of load average in Linux.
On Thu, Apr 26, 2012 at 8:18 PM, Lists lists@benjamindsmith.com wrote:
Problem isn't so much actual "speed" but causing network monitors to freak out due to "high" load average when performing backups.
So don't plug the USB into the server. Put them on some other machine and run the backup over the network. Network interfaces normally have some intelligence.
Seems silly that load average would climb to 2.x or more copying some files on an otherwise lightly loaded server.
Load is the count of processes waiting for CPU times. If you use an interface that takes all the CPU cycles to manage, what do you expect to happen?
On 27/04/2012 02:18, Lists wrote:
Problem isn't so much actual "speed" but causing network monitors to freak out due to "high" load average when performing backups. I can make exceptions for servers doing backups, but then I don't get notifications when the load is legitimately high. I can make exceptions only during backup times, but that increases complexity.
When using nagios or zabbix you're better off working out what the average load is over a large period of time (say, day, or week) and then comparing the load over the last 15 miuntes, say, to the average load. This should even out spikes, and also be comparing to a good base level load.
On Fri, Apr 20, 2012 at 12:42 AM, Lists lists@benjamindsmith.com wrote:
Problem as follows:
Plug in an external USB drive.
Mount it anywhere. Doesn't matter how.
Copy a few GB of data to the drive from a non-USB disk.
Watch the load average "climb" to 5.x, sometimes 10.x or more. Why?
This on an otherwise unloaded system. Doesn't matter how many cores, how much RAM, 32/64 bit, etc.
Why should copying some files to a USB drive cause load averages to climb so high? (and network monitors to freak out?)
The CPU has to do the work of the transfer over usb - which is why it is cheap. Real disk controllers use DMA without a lot of CPU involvement.
On Friday, April 20, 2012 10:54:51 AM Les Mikesell wrote:
The CPU has to do the work of the transfer over usb - which is why it is cheap. Real disk controllers use DMA without a lot of CPU involvement.
And this includes USB 3.0, incidentally. I have found that on my Fedora 14 (soon to be C6) laptop with a USB 3.0 ExpressCard interface that the load is less on USB 3 and it is on the laptops built-in USB 2 ports.
YMMV.
-
Giovanni Tirloni -
Lamar Owen -
Les Mikesell -
Lists -
Shaun