Support

mistwang · #11 09-02-2008, 08:27 AM

From the CPU cycles consumed by the two lshttpd processes, two processes has been well balanced, so the configuration should be all right.

And it does look like the kernel allocate I/O operations to only one CPU core only, it is up to the kernel I/O scheduler, not much we can do about in user land applications.

It might be at the peak I/O throughput for the disk array, since those are random read accesses. there are a few things we can improve if have not done yet.

Make sure NCQ has been enabled for the RAID card and hard disks. try a different I/O scheduler in the kernel.

We can try a 4-CPU or 8-CPU license on this server, see if it get any better when more I/O worker processes are used. It may help a little with NCQ enabled I think.

IrPr · #12 09-02-2008, 03:17 PM

Quote:

Originally Posted by mistwang

From the CPU cycles consumed by the two lshttpd processes, two processes has been well balanced, so the configuration should be all right.

And it does look like the kernel allocate I/O operations to only one CPU core only, it is up to the kernel I/O scheduler, not much we can do about in user land applications.

It might be at the peak I/O throughput for the disk array, since those are random read accesses. there are a few things we can improve if have not done yet.

Make sure NCQ has been enabled for the RAID card and hard disks. try a different I/O scheduler in the kernel.

We can try a 4-CPU or 8-CPU license on this server, see if it get any better when more I/O worker processes are used. It may help a little with NCQ enabled I think.

WOW! AWESOME!

switched IO scheduler to deadline and some other tips with kernel, now server performances increased dramatically

this mrtg graph indicates that:

Look output is about 3x now

Thanks George, you are the man!