Re: Number of IOs Per Second

"Connor McDonald" <connor_mcdonald_at_yahoo.com> wrote in message news:411A1C97.7D97_at_yahoo.com...
> CHANGE USERNAME TO westes wrote:
> Your queries won't generate "IOs per sec", they will simply generate
> "IO". You could imagine a worst case scenario would be a nested loop
> cartesian join (8m x 8m x size of table A x size of table B). eg 200
> bytes per table yields around 2 terabytes of disk read. Hopefully your
> queries will be better than that :-)
>
> Once you've got a handle on typical IO, and how long the queries are
> allowed to take, *then* work out what IO's per second you need to
> achieve the results you need.

That's very clear thinking, and your example for calculation of amount of data is helpful.

What should I be using as a figure for the amount of data in each I/O? I am referring to disk-level I/O and not OS-level I/O, since I have as one data point that a single physical disk will give me something around 100 disk-level IOs per second.

So in your scheme, I am left with a formula to calculate the number of disks I need based on amount of data I collect and time I need to collect it.

# of disks = Y GB per data collection / [ 100 IOs per disk per second * .000256 GB per IO (or whatever the physical IO size is) * Y Desired Seconds per data collection ]

Is this the basic idea?

-- 
Will
westes AT earthbroadcast.com