Solaris

If you want to debate the accuracy of anything in this thread, do so in this thread.

If posts are not related to the original topic they will be deleted.

Are you using ZFS on this machine?

Glad you asked about ZFS, I was about to do the same.

The OP has 8 GB of RAM, 1.5 GB is used by processes while 6.5 GB is used by the kernel. This per se isn't a problem. Unused RAM is wasted RAM anyway.

One major change ZFS introduced compared to UFS is the page cache is not used to cache file content. Kernel memory is used instead . This dramatically increases the kernel RAM usage metric but has no real consequence as this cache memory is still free memory from a system's viewpoint.

On the other hand, if the OP isn't using ZFS, then there is a problem to investigate further. It is particularly a problem because the kernel seems to be already larger than the swap space and might grow even bigger before a potential panic. Assuming the dumpadm intricacies are unknown to the system administrator, a crash dump might be truncated and then be partially or, worst case scenario, wholly unusable.

The reason why I was picky about that point in previous postings.

Yes, indeed that was the reason for asking. The behavior you describe is was is expected from the ARC cache, however it is worth noting there have been a number of bugs related to the cache memory not being freed quickly enough, so it doesn't always work the way it should resulting in unnecessary paging. I have actually run into this issue, but I believe I have only seen it on x86 machines. It is possible, and advisable, to limit the ZFS cache for this reason if there is a possibility that it will interfere with other operations.

BTW,

I was reviewing the docs on the ZFS filesystem and dumping and found this example:

Using a ZFS Volume as a Swap or Dump Device (Solaris ZFS Administration Guide) - Sun Microsystems

Code:

# dumpadm
      
Dump content: kernel pages
Dump device: /dev/zvol/dsk/rpool/dump (dedicated)
Savecore directory: /var/crash/t2000-bdr-02
Savecore enabled: yes

Beware that this is beta documentation showing a feature AFAIK not yet released in Solaris 10.

8.ZFS Advanced Topics (Solaris ZFS Administration Guide) - Sun Microsystems states:

While Zvol can be used, a plain file on ZFS as a dump device is not supported even with Solaris Express, and there is no plan to implement it.

1. mdb -k shows this to me :

Kernel --- 851120 ---- 6649 ----82%
Anon --- 99544 ---- 777 ----- 10%
Exec and libs ----- 8788 ----- 68 ----- 1%
Page cache ----- 24448 ----- 191 ----- 2%
Free (cachelist) ----- 32542 ----- 254 ----- 3%
Free (freelist) ----- 25260 ----- 197 ----- 2%

Total 1041702 8138
Physical 1024847 8006

2. We are using Vxfs.

3. I dont have memtools installed.

Q1. Output of "mdb -k" which shows 82 % Kernel , Does that mean that my kernel code is taking that much memory ( I dont think so ) OR does that mean that Kernel asked for this much memory for File , directory cache + buffers ( i think so )

Q2 . If kernel has asked for this much memory for file caching then why on running some other applicaiton it starts doing page outs which is like shortage of RAM? Ideally RAM used for file caching should be given to any program which want to get executed in RAM.

Is there any tool for checking file cache in solaris 10 except from memtools ?

For ex in linux we i can chk using "free" about file cache , directory cached and buffers .. and i havent seen free in solaris although i think the memroy management for solaris and linux are almost similar ?


Thanks in advance
Raj