We use cacti to view system usage on our machines. Cacti shows that almost all memory is used by the system. We have 96GB of RAM in one machine and only one SAP system running on that machine. top shows the same amount of RAM usage as cacti. But someone told me that the hp-ux 11.31 os is not letting go of the memory before it is used. How could i know how much memory is really used? I have just moved this SAP system from a 11.23 machine with 40GB RAM, and it was fine on that. Why is top showing that almost all memory is used? It is the same on my other new 11.31 machines.
i only say so because everyone is talking about swap should be double the memory. But i do not understand why, because you do not want the machine to swap at all.
How do i read the swapinfo -tam right? to understand if memory is under pressure? As i said before the top command shows only 3,7GB of free ram.
>i only say so because everyone is talking about swap should be double the memory. But i do not understand why, because you do not want the machine to swap at all.
Exactly. If you read similar issues in the forum you'll see the experts poo-poo that double the memory cliche. Especially if you have 100 Gb.
>How do I read the swapinfo -tam right? to understand if memory is under pressure?
swapinfo will not show anything related to physical memory usage. It will display the swap space details. The memory column in swapinfo is not the physical memory usage which people misinterpret. It is pseduo swap which is part of physical memory.
If you really want to know the physical memory usage use kmeminfo tool or glance.
The old (archaic) rule of swap should be 1 or 2 x RAM is outdated.
By the way, it is a bad idea to have multiple paging devices on the same physical disks:
dev 8192 0 8192 0% 0 - 1 /dev/vg00/lvol2 dev 48000 0 48000 0% 0 - 2 /dev/vg00/lvol10
I do not know how many disks are in VG vg00, but I hope your two paging devices are not on the same PVs. That is one of the basic rules when we teach HP-UX System and Network Administration courses (SNA I and II).
Since you use HP-UX 11.31, I would only have primary swap (small, 4 or in your case 8 GB)... The idea is not to swap ever :)
v3 is really no different than v1 or v2 in this respect. HP-UX historically favors caching for performance -- if there's no memory pressure, memory is allowed to remain in caches instead of being fully "Free" -- hence you can expect free memory to go down to and linger around 1%.
Given your filecache_max is the default 50%, and assuming filecache_min is also the default (5%) -- that's 45% of reservable memory (95% of RAM) likely in this cache if there isn't much load. [This is where kcusage or kmeminfo details are helpful... what amounts does it show for User? Kernel Dynamic? UFC?]
Your swapinfo output does show little in the memory swap category -- so your kernel dynamic shouldn't be very large. Hence, my suspicion that you're just using all of the file cache. Which is fine -- you'll get that memory back when and if you need it.
Just as a side comment, 5.19 is really old. That's likely why the User result is pretty obviously garbage (I seriously doubt there's no processes running, including kmeminfo itself). This also predates UFC, which is why you don't see where most of the memory is [kcusage shows that UFC is at 49011838976 or 45.6Gb]. The arenas contributing to kernel dynamic (which is pretty low as expected) are file/process related mostly (Btrees are sub-region VM metadata, pfn2v is just a per-page cost and unrelated, VxFS and Fcache is kind of obvious). Beyond that -- the Static costs are unavoidable, so don't worry about them.
Personally, I'd say leave the system alone -- it is working as expected and you'll get the memory back from the file cache when you need it.
If there's some pressing terrible reason you really want to see more free RAM you can reduce filecache_max to be closer to filecache_min -- forcing UFC to release the memory back to the system.
This has been a great thread for me. Really useful! I will leave the system alone... :) I just think it´s sad that i do not know how much more i can push the system. For example if i want to run another SAP system on this one. The only way to see if it can handle it, is to start SAP and see if the machine starts swaping?
No -- you know that you have 3Gb laying around idle that another SAP could use... and you have 45% of memory (50% max - 5% min) which SAP can contend for. That won't cause paging [swapping is when processes are deactivated and should only happen under critically low memory situations] per se -- if the file-backed pages are unmodified, they'll just be dropped. If they were modified, they have to be sync'd eventually anyway.
A trivial way to test the effect of raising your workload by N Gb is to write a program with mlock privileges, allocate N Gb and mlock it -- then look at swapinfo (is any swap actually used), po in vmstat (may get a spike since UFC pageouts count, but should settle down), etc. But in essence -- the memory between max and min is available and fairly easily reclaimable for new workloads. No promises that less memory available for filecache won't affect performance of the system aftwerwards, of course if there's a lot of file I/O involved in said workloads.
