Re: when to stop buying RAM & disk

Terence Kelly <tpkelly@mynah.eecs.umich.edu> wrote:
> Is there a literature, folklore, or accepted wisdom regarding how to
> trade memory expenditures for savings in external bandwidth and/or
> client latency?

Well, "common sense" would indicate that you should pick a timeframe,
approximate the savings over that timeframe, and compare that to the costs
to realize that savings. So spending $500 on memory which saves you $100
in bandwidth over the next year doesn't make much sense. Of course,
spending $500 on memory which saves you $500 in bandwidth over the next
year _also_ doesn't make much sense.

I'd actually hesitate to bother spending the money unless you estimate it
would save at _least_ twice as much over the timeframe, because most likely
your estimate of savings is not precise, and your estimate of current costs
is not precise (the memory might only cost $500, but you also spend $100
getting it from the vendor into the machine, say).

As far as client latency... savings in external bandwidth are a pretty sure
bet, because it's hard to imagine ways where IMS requests will cost you
more than straight GETs. But having a proxy cache in the middle might
actually _decrease_ overall client latency, even with high hit rates.

Later,
scott
Received on Tue Jan 11 2000 - 12:19:54 MST