[AusNOG] Question about hardware spec for a DC

[Paul Gear]

Good point.  Because you can never be sure that the remaining PS is not 
going to fail, it needs a maintenance window regardless.  I would expect 
that the probability of a PS failing would go up if its load is 
dramatically increased, which is exactly what would happen if the other 
PS input was removed.

Paul

On 04/24/2014 12:15 PM, Colin Stubbs wrote:
>
> I once thought like that too.
>
> But if you're re-organising power and yanking cables out and plugging
> them back in on anything other than a lab system you should be working
> within a declared maintenance window anyway.
>
> There's too much risk that something will come out that shouldn't, or
> that something will shift just enough to cause a problem.
>
> It might be acceptable to do it without a window for a UAT/test/dev
> system but anything revenue generating or revenue supporting... hell no.
>
> Having the convenience and ability to re-arrange power cables at any
> time of day without risk and loss of $'s stems from designing systems to
> cope with failure to start with.
>
> That may in small part be because you put two PSU's in the boxes; but it
> won't be the only reason.
>
> -Colin
>
> On 24 April 2014 11:30, Paul Gear <ausnog at libertysys.com.au
> <mailto:ausnog at libertysys.com.au>> wrote:
>
>     In my judgement, the big benefit of dual power supplies
>     (particularly on servers, switches, & routers) is that it enables
>     reorganising of power circuits in racks without requiring a downtime
>     window.  Like you say, a comprehensive view of high availability
>     means designing N + 1 redundancy at the system level, but dual power
>     supplies to those N + 1 devices is a big convenience factor.
>
>     Paul
>
>
>     On 04/24/2014 11:25 AM, Colin Stubbs wrote:
>>     IMHO
>>
>>     Avoid that auto power-transfer stuff in rack if you can. Those
>>     devices are best used only for low end boxes like NTU's/etc with
>>     which you can only ever install/utilise a single box at a time.
>>
>>     Buy equipment which has dual PSU's as an option in preference, but
>>     don't mandate it or mandate buying the two power supplies.
>>     Mandating it will just mean you're wasting part of your budget
>>     unnecessarily on every single purchase as you may force yourself
>>     to buy bigger boxes than required; and/or buy more PSU's than you
>>     need.
>>
>>     Avoid *depending* on dual PSU's if you can, e.g. don't design
>>     anything with the assumption having two power supplies in a box
>>     will keep everything working if there is a loss of power or if one
>>     of the PSU's fails.
>>
>>     Design for failure with N+1 redundancy at a system level, e.g.
>>     install two or more of every box and use them in active/active or
>>     active/standby capacities. If you do that you won't necessarily
>>     require two power supplies in each to achieve a very high level of
>>     availability.
>>
>>     Choose to use dual PSU's primarily based on the location and the
>>     power infrastructure available, in combination with how many boxes
>>     you have. e.g. if you're in a crappy DC where they can't deliver
>>     access to two genuinely independent sources of power the value of
>>     having two PSU's is greatly reduced regardless of how many many
>>     boxes you've installed.
>>
>>     Read up on HA concepts if you're not sure what you need or why.
>>
>>     13 years old now but this book is still handy and the concepts
>>     still hold true,
>>
>>     http://www.ciscopress.com/store/high-availability-network-fundamentals-9781587130175
>>
>
>
>     _______________________________________________
>     AusNOG mailing list
>     AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
>     http://lists.ausnog.net/mailman/listinfo/ausnog
>
>