Re: Copy slows down; is it normal?

Hi Bennett,

Thanks for the heads-up, but I don't believe the multiplicity of the processors is a problem in my case. SMP is just another way of allocating virtual machines. All the time since i486, unix systems did use one VM per process, but now that you have SMP, there are chances that concurrent VMs will run in more than one physical, which does not increase the per-process performance but allows you to run more than one process without time sharing, which is great for database and web servers.

> As I know, Linux kernel 2.0.X(RedHat 5.1) does not directly support DUAL
> CPU/ SMP.

It does, but for some reason, this option is not configurable through 'make xconfig' or 'make menuconfig' menus. And it is not built into the default kernel. Setting SMP = 1 in the top level Makefile does it.

> Therefore your system may work as a single 400MHz i586 system
> only.

>From each process' standpoint, it always does, unless the process fork()s. Performance-wise, it looks like a single processor machine for each postgres backend, but it appears multiple for the postmaster. The only difference for the backends while being run on a dual machine is that they do not compete with each other when there are only two of them, or compete less when they are many.

> I heard that Linux 2.1.X support SMP but it is still a developer
> version.

That's what the note in the makefile has always been saying:

# NOTE! SMP is experimental. See the file Documentation/SMP.txt
#

I arrogantly ignored this note for over two years now (since RedHat 4.1, I think). I used SMP in more than a few dual machines, including the one I am sitting at now, without any problems so far. I believe the feature is worth recommending it to others.

Sincerely,

Gene