Re: [GENERAL] advice on buying sun hardware to run postgres

> Bruce Momjian wrote:
>
> > When you get above two processor, the SMP method used by Linux, FreeBSD,
> > BSD/OS, etc do not work as well as a kernel-level SMP like Solaris.
>
> "kernel-level"? All SMP is kernel level. Solaris does much better but
> there is no fundamental architectural difference as far as I know.

There is. Linux/*BSD use only a big kernel lock, or several big kernel
locks. Solaris allows multiple CPUs/threads to run in the cpu at the
same time. This ability requires quite a bit of overhead, which make
Solaris/Slowaris slower in many operations.

I am no Solaris apologist, and don't prefer it myself, but it is
important to understand what they are trying to do vs. Linux/*BSD.

>
> > In the Linux/*BSD SMP, you can only run a kernel call on
> > one processor at a time.
>
> If that was true, why has Linus just spent 2 years making kernel locks
> more fine grained? If you could only run one kernel call at once anyway
> then they must have wasted their time eh?

He is increasing the number of locks in the kernel, making it more
fine-grained. Still probably does not have the same capability as
Solaris, where kernel threads are very finely grained.

>
> > IDE does not allow multiple outstanding request, can not access multiple
> > disks at the same time, and requires more CPU to do the transfer than
> > SCSI. This last point is normally forgotten.
>
> Surely IDE can handle multiple disks at the same time if they are on a
> different controller.

Yes, but few controllers allow unique IRQ's, at least in the past, and
you can't have more than 2, no? It is a kludge.

--
Bruce Momjian | http://www.op.net/~candle
maillist(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026