Performance issue.

[rajeshk]

Hi,

I am having the following configuration.
Problem: While running 3 compilation simultaniously both the processor utilization is 100%. Whether i need to upgrade RAM or i need to go for new machine configuration like K class server. Because i need to run 6to 7 compilation in the next few months.

#model
9000/800/L2000-36
#dmesg | grep "Phys"
Physical: 2097152 Kbytes, lockable: 1554408 Kbytes, available: 1792672 Kbyte
s
#swapinfo -a
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1048576 130348 918228 12% 0 - 1 /dev/vg00/lvol2
dev 2097152 131228 1965924 6% 0 - 1 /dev/vg00/lvol9
dev 2097152 130672 1966480 6% 0 - 1 /dev/vg00/lvol10
reserve - 1101300 -1101300
memory 1558880 696356 862524 45%

[Ravi_8]

Hi,

you are using L-class whic are superion than K-class servers.
fro the info you have 2GM RAM and 5GB swap.
excess of swap causes excessive I/O. You could try by reducing the swap by 1GB or increase RAM by 512MB ( I prefer reducing swap by 1GB)

[Frank Slootweg]

Compilations, especially with optimization, are very CPU-intensive, so seeing 100% CPU utlization is to be expected.

How many CPUs do you have?

As to memory:

Do you have Glance? If so, what are the "Mem Util" percentages during the compiles? If not, what is the "memory free" number reported by "vmstat 1 10" during the compiles? (The vmstat "memory free" number is in 4KB pages, i.e. you can calculate the memory utilization percentage.)

[Shannon Petry]

Swapinfo does not show memory use in the true sense, only memory used for swap.

Looking at the output from your swapinfo, I'd say your having more of a CPU bottleneck than memory. Swap will start to get busy as memory gets full. Your swap is not highly used at least in this ouput.

Look at top, and see what the memory used really is. If your 100% for long periods of time, and need to double your workload then at a minimum you should double your RAM.

However, you should also look at your CPU usage. If your CPU's are loaded, then the additional memory alone wont do much good.

Regards,
Shannon

[Bill Hassell]

As mentioned, compilations are CPU intensive so it is expected that each compile will use about 100% of one CPU. If you need to run 6 or 7 at the same time, you'll need 6 processors to allow each compilation to proceed at full speed.

That if course assumes that there is enough RAM to keep everything in RAM. If not, then processes will be deactivated and paging to the swap area will occur. That will impact compile jobs making them as much as 100x slower. You can see if paging is occurring with vmstat (look at the po column). If po is dozens to hundreds, then you need a lot more RAM.

The L-class is much faster than the (no longer manufactured) K-class. A 6-processor L-class with 4 to 12 Gb of RAM should allow compilations to complete with little delay. If the completion time for the compile jobs is not important, you can run on just a couple of processors and a few hundred megs of RAM. The impact will be enormous but the system will still work, albeit slowly.