Originally Posted by Mara
The post above was moved from http://www.linuxquestions.org/questi...d.php?p=216358 because of two reasons: the orginal threads is very old, and, what's more important, I have a stron reasons to think that the problem mentioned in the thread above is completely different from the one discussed in the thread it was orginally posted.
alpha754293, the problem you have is with the kernel resonses (the timeouts you have mentioned). If they're not requesting much more RAM than the machine has, the problem is not in the amount used. How do stats look like? The most interesting is the result of 'free'. What does it show? And also, why do you think it's a memory access problem?
Well, right now the system is reporting 1007 MB (according to the system monitor) in RH9, of which, 997-998 MB of it is used (representing 99% of all available memory).
I did a full install (i.e. install everything in the options in RH9), about 4.8 GB in all.
I still haven't been able to come up with the same number that is being reported as being used checking the process and thread memory usages, which begs the question: "what gives?" (hence the relevance to the old thread.) I also figured that rather than starting a new thread about the same question/issue, and cross-referencing it, that I just posted it where I thought it belong.
No matter what I do with the system, it pretty much around that level.
I have frozen the package manager numerous times and also crashed nautilus, gnome, mozilla.
I know that this isn't a faulty system (CPU, memory, mobo) because prior to the server migration, it was running Solaris 10 3/05 stable on the same platform. (I switched because I wanted HW ATA RAID5 support, and the Promise SX6000 is not supported there.)
Free mem is around 10 MB. (I do not know how to check it via CLI.)
Based on my experiences with AIX, Solaris, and HP-UX, this is highly unusual/abnormal.
I also do not agree with the person that responded saying that the way how Linux is managing the memory is better. If the idea is that it is going to index it; then I would expect much better response times than what I have been getting on the system.
If there is significant activity on the array (which I am also not accustomed to the lack of being able to monitor the status of the drives); it usually takes a while for the system to kick in, even if it's something as simple as 'ifconfig -a'.
From my experiences with Solaris 10, I can push/pull 17.5 MB/s across the system and the system would only be using about 100 MB or so of RAM, and wouldn't have issues processing other commands.