Hi,

I'm about to start a website on a shoestring, so I'm looking to get
the most bang for my buck in terms of hardware. I will be running this
site on a linux cluster with mysql and there will be lots of reads
from the DB and not so many writes.
For various reasons, I have to create my own 'datacenter' so can't use
other provider so please don't suggest that :-). I'm thinking I'll
start with maybe 5-6 or so machines and add to it as necessary.

So has anyone done studies on what is most cost effective? f.ex. on
the processors a Xeon 1.6 quad-core costs about 10 times more than a
AMD Sempron 3600+.
But the Xeon does have 2MB L2 cache compared with the AMDs 256kb and
I'm not sure exactly how much faster it is (is it 10 times faster? I
think not).
How important is the L2 cache in DB applications?

Also, the bottleneck with probably be the disk, so would it not be
better with lots of cheap-cpu machines compared to a few high-end-cpu
machines? Since there will be mostly reads from the DB I would think
it better to spread it on many smaller machines. So would 10 small
machines not be better than 5 larger? (assuming the larger ones have
f.ex. dual or quadcore cpu's and the smaller have semprons for
example). There would be more disks in the low-end-cpu machines since
there are more machines, so wouldn't that be better for the
bottleneck?
What about motherboard? is there a lot to gain from a high-end board
compared to a cheap one? is it worth the extra costs for this
application?

Finally, I have no idea if I should start with 5 machines or 20? or 2?
does anyone have some real-world examples? Like f.ex. what did myspace
start with? or facebook? I know I will have to run some tests and
experiment, but it would be nice to have some idea of what other DB
sites started with.

My gut-feeling is to go with cheap cpu's (like AMD sempron 3600+),
lots of fast RAM, cheap motherboards, very good gigabit NICs and
switch, fast high rpm high cache SATA disks (500gb or so) (one in each
server), cheap tower cabinets. And then start with maybe 5 or 6 of
them or so. My only worry with that setup is the small L2 cache. I
wouldn't go with redundant PSUs or disks (raid) or NICs because there
is redundancy in the cluster itself anyway. What do you think of that
setup?

Thanks.
Tobias