RE: odd memory behaviour in 2.4.15

RE: odd memory behaviour in 2.4.15

[Adam Margulies]

I know this has been covered before, however I
am puzzled by this situation and so am asking the list for help.

I have a rack of 14 linux machines, each with 4GB of RAM.
They run great, no problems, using the 2.4.10 kernel.

I have a separate linux machine with identical hardware
running 2.4.15 which also ran well, until I upgraded its
memory to 4GB. Then it ran 3 times slower than with 1 GB of RAM.

I know the solution is to put an "append="mem=3900M""
in lilo.conf, sacrificing 287MB in the process,
but my question is why do I have to do this?
and why do the 4GB 2.4.10 machines run fine without it?

I've compared their .configs and while they are different
they aren't different in the important settings, such
as the memory configuration settings. So I don't think
it is a kernel config issue.


_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp

Re: odd memory behaviour in 2.4.15

[Alan Cox]

> I have a separate linux machine with identical hardware
> running 2.4.15 which also ran well, until I upgraded its
> memory to 4GB. Then it ran 3 times slower than with 1 GB of RAM.

We end up copying data around a lot more with the 4 or 64Gb kernels. Some of
this unneccessarily. It may be down to your workload it might be a vm
quirk. Do the boxes all run the same workload.

Jens Axboe has patches for some setups that basically kill the slowdown
for all cases providing your I/O controller is sane