[nycbug-talk] green monster server

[Charles Sprickman]

On Thu, 28 May 2009, Miles Nordin wrote:

>>>>>> "cs" == Charles Sprickman <spork at bway.net> writes:
>
>    cs> I'm looking for some input on how to make it less power
>    cs> hungry.
>
>    cs> most power supplies are lucky to be 70% efficient).
>
> no, there's an 80plus logo consortium now.  This means the power
> supply is over 80% efficient even when running at much less than full
> load.  most but not all supplies on newegg have the logo.

Excellent.  I got the 70% figure from some article I read a few years ago 
about a push for a "new" power DC power standard for datacenters that 
would run either 12V or 5V to cabinets rather than AC or 48V DC.  The idea 
was to basically eliminate the waste factor inherent in every server in 
the datacenter.  The power savings were astronomical.  Chances of this 
thing being standardized and implemented seem minimal though...

Is there any way to determine which power supplies will draw the least 
power when not actually loaded up?  Or is part of that 80% initiative to 
mandate that even if the box is idle and pulling like 40W the PS is not 
pulling 100W out of the wall?

Just nuking this cheap-ass triple-redundant sham will surely bring me 
under 150W at idle I suspect/hope.  I can live with that power draw for 
now.

>    cs> 8 250GB ATA drives
>
> 2 or 3 1TB drives instead?

Eventually.  I got this for exactly $0.  No budget beyond maybe $80 for 
the power supply.

> i think dual parity is a good idea though.

I have much reading to do on ZFS.  All I've seen are the gee-whiz 
writeups.  Now that FreeBSD has imported the latest and there are more 
committers involved I'm ready to learn and play.

>    cs> Too many fans, at least 10.  I'll remove some until I see the
>    cs> temp getting too warm for the drives.
>
> this case is pretty quiet:
>
> http://www.servercase.com/miva/miva?/Merchant2/merchant.mv+Screen=PROD&Store_Code=SC&Product_Code=CK4020&Category_Code=MS

It's in the garage, so noise is no problem.  The frontend will likely be a 
$250 hackintosh running Plex (http://www.plexapp.com/).  I have no cable 
TV or satellite, just Netflix and BT for video entertainment and a sizable 
library of DVDs to rip.

>    cs> As I can afford it, I'll retire drives and put in larger units
>    cs> to bring the drive count down (see ZFS above - hopefully it's
>    cs> magical enough to allow this type of tinkering and migration).
>
> nope.

crap.

> you can't decrease drive count without destroying and recreating the
> pool.  The only exception is to remove a mirror component---that you
> can do.  You can increase the size of individual drives obviously, and
> also if you increase the size of all the drives in a stripe the pool
> will get bigger (proportonally to the smallest drive in the stripe).
> but there's no way to decrease the size of a drive, nor to migrate
> data off a drive and remove it.

Not even remove it?  That's weird.  So if you put together an array today 
with 1TB drives and 5 years from now you are stuffing 5TB drives in, 
you're SOL?

I could probably start with fewer drives, leaving some slots open.  When I 
get the 1TB drives, I could make a new pool there and at least copy the 
data over.  Anyhow, it should be fun to figure out.

> they are working on this capability.  they call it bp rewrite i think.
> but they've been working on it for more than a year.

I'll google around a bit.

Thanks,

Charles