Scott Marlowe wrote:
> On Tue, Oct 6, 2009 at 8:26 AM, Scott Marlowe <scott(dot)marlowe(at)gmail(dot)com> wrote:
>
>> On Tue, Oct 6, 2009 at 7:21 AM, Nikolas Everett <nik9000(at)gmail(dot)com> wrote:
>>
>>> If my un-word wrapping is correct your running ~90% user cpu. Yikes. Could
>>> you get away with fewer disks for this kind of thing?
>>>
>> Probably, but the same workload on a 6 disk RAID-10 is 20% or so
>> IOWAIT. So somewhere between 6 and 12 disks we go from significant
>> IOWAIT to nearly none. Given that CPU bound workloads deteriorate
>> more gracefully than IO Bound, I'm pretty happy having enough extra IO
>> bandwidth on this machine.
>>
>
> note that spare IO also means we can subscribe a slony slave midday or
> run a query on a large data set midday and not overload our servers.
> Spare CPU capacity is nice, spare IO is a necessity.
>
>
More importantly when you run out of I/O bandwidth "bad things" tend to
happen very quickly; the degradation of performance when you hit the IO
wall is extreme to the point of being essentially a "zeropoint event."
-- Karl