topic Re: Very high inact_dirty pages on RHEL AS 3.0 in Operating System - Linux

Very high inact_dirty pages on RHEL AS 3.0

Alison Ayson — Thu, 15 Sep 2005 14:09:24 GMT

We are running RHEL AS 3.0 (U5) on an HP DL740. This server has 20 GB of memory but with less than 470 megs free.

I know that it's "normal" to see most of the free memory cached, but is it normal to have such a high number of inact_dirty pages?

The system seems to be suffering when the inact_dirty pages gets so high, and we end up rebooting to clear it up.

I understand that kflushd is supposed to clear out these pages, but this is not happening.

Has anyone ran into this problem? if so, what did you do to fix it (assuming you could fix it) besides rebooting?

Here's my /proc/meminfo output:

total: used: free: shared: buffers: cached:
Mem: 21049479168 20561117184 488361984 0 82972672 19616972800
Swap: 10737147904 0 10737147904
MemTotal: 20556132 kB
MemFree: 476916 kB
MemShared: 0 kB
Buffers: 81028 kB
Cached: 19157200 kB
SwapCached: 0 kB
Active: 1013988 kB
ActiveAnon: 494464 kB
ActiveCache: 519524 kB
Inact_dirty: 14279764 kB
Inact_laundry: 3850932 kB
Inact_clean: 587412 kB
Inact_target: 3946416 kB
HighTotal: 20053948 kB
HighFree: 460028 kB
LowTotal: 502184 kB
LowFree: 16888 kB
SwapTotal: 10485496 kB
SwapFree: 10485496 kB
CommitLimit: 20763560 kB
Committed_AS: 589112 kB
HugePages_Total: 0
HugePages_Free: 0
Hugepagesize: 2048 kB

Re: Very high inact_dirty pages on RHEL AS 3.0

Ivan Ferreira — Thu, 15 Sep 2005 15:24:11 GMT

From here:

http://www.redhat.com/advice/tips/meminfo.html

I see that Inact_dirty: Dirty means "might need writing to disk or swap." And you are not using swap space.

The word "might" is the key. So, as almos all memory is cached, it sould be not a problem. The kernel "might" think that need to be writted.

You can try a simple sync, to see if there are files that need to be written to disk.

Re: Very high inact_dirty pages on RHEL AS 3.0

Alessandro Pilati — Fri, 16 Sep 2005 07:37:32 GMT

Hi Alison,
Inact_dirty memory is made by data that waits to be written to disk or swap,
so if you run 5-6 times the "sync" command,
you should see a decrease of that value.

Let us know...

Cheers,
Alex