UNIX for Dummies Questions & Answers

Hello all,

I have a sunfire 280R with 2 750 MHz CPU and 2 Gig of RAM. But when I compile things it does not seem to be using all of it's resources thus the compile take forever. At most it seem to be only using at most 10% of CPU and 5% of swap sapce. I have tried to renice the pid of the compile, but it still does not seem to do anything. Does anyone know what I can do to tune/increase CPU usage? THANKS all

You may want to use sar, vmstat, top, or some other utility to determine what the bottleneck is on your system. My guess would be you are waiting on I/O to disk. You could increase your CPU usage slightly by using faster disks, although I doubt it is going to change your situation much. I don't do a lot of compiling but I am guessing it is a fairly I/O bound process and is probably bound to 1 CPU which means you will most likely never use both CPUs anywhere near 100% while compiling. If you don't have anything else running on the system, the CPU's are so much faster then anything else that unless you are doing some heavy crunching, the CPU will always be waiting for disk, memory, etc.

Hi ,
I am facing a similar problem on the same configuration.
In my case the problem is the cpu usage does show 97 - 99 % usage and also there is no i/o wait, when i fire multiple processes.
but i don't get the performance.

Basically i m running a application which generates small files.I need to generate around 170 files per minute. If run just one process one file takes 5 secs to generate. When i run say 10 process it shows 40 secs to 200 secs for certain files and also the average time is around 4 secs per file.

Is it that i m overrunning the process or do i need to set some kernel parameters.Please help.

Zing

Generating files is an expensive process and it's not a process that is cpu bound. Using ten disk drives would be more useful that 10 cpu's.

Also how many files in the directory. Creating file number 1001 in a directory costs a lot more than creating file number 2. Keep the ratio of files per directory as low as possible.

Look at the mounting options on the filesystem. Metadata is typically synced rather aggressively. There may be a way to give up reliability to get performance. For example with a vxfs file system on hp-ux you could set minchache=tmpcache.

Thanks TioTony,

I was using top to look at the cpu usage. I have also used sar, vmstat and prstat -a to look at things. All show the same thing, the CPU and swap sapce is not being used. There is nothing else major running on the system. I have 10,000 RPM disks. Let me add one thing to this, the home directory is mount from another machine. To my understanding this should not be an issue. Any input thanks.

Are you compiling to the home directory or using any files in it during the compilation process? What was the waiting for I/O percentage with the utilities you used. What is the load average? 10K disks are fast but no disk (even a RAM disk) is likely to be as fast as your CPU. I still have the feeling your bottleneck is I/O. If you are reading/writing to the home directory, you may want to consider moving the files local, compiling, and then moving the resulting files to the home directory. This may help utilize the CPUs better. If your load average is high, then you may want to shut down non-essential programs, at least while you are compiling.

Also check out the stats on the system serving out those home directories since it looks like you are using them when compiling. You may find your problem there.