[HCoop-Discuss] VPS only looks cheap compared to NYC

[Adam Chlipala]

Adam Megacz wrote:
> Adam Chlipala <adamc at hcoop.net> writes:
>   
>> What happens if some component of your hosted server fails?  How long 
>> does it take to get the part replaced, and how much does that cost?
>>     
>
> I buy extremely reliable hardware.
>
> In a situation like hcoop's, I would advise having a spare machine
> online at all times.
>   

You can probably tell that I'm pretty ignorant about the nuts and bolts 
of hardware and hosting today.  No one else is coming forward to keep 
this discussion moving, so I'm going to keep pushing on this, and I hope 
you can be tolerant of my ignorance. ;)

I've heard a lot about high disk failure rates, and it sounds like 
you're saying that disks are the only kind of hardware that we need to 
accept as having failure rates high enough that we need to contemplate 
dealing with failures at anything but a leisurely pace.  Am I 
interpreting you right?  Does everyone else on the list who knows about 
this kind of thing agree?

Also, what makes two the magic number of machines to have available at a 
time?  (I'm quite willing to accept that doing the calculations about 
component failure rates leads to that conclusion; I'm just curious if 
this is accepted wisdom, or if some folks would recommend 3 machines, etc..)

Adam Megacz wrote:
> For everything other than hard drives, I don't recommend buying
> commodity components.  Instead, buy a brand known for reliability (IBM
> or Sun), but get hardware that's "one generation old" on the used
> market.  The performance hit isn't that bad, and the reliability is
> more important.

Does anyone else watching this thread have a different opinion on the 
complete set of vendors whose hardware we might want to use, for 
reliability or other reasons?

Adam, which model numbers of IBM and Sun servers do you think are the 
"sweet spot" choices to buy right now, and where did you get such a good 
price on your beefy IBM server? :)

Do you think that the KVM feature of the service processor that you 
mentioned before is enough that we should prefer IBM over Sun overall?

(Sorry for all the questions.  I hope you don't mind sharing your 
wisdom. :])