[nycbug-talk] green monster server

Miles Nordin carton at Ivy.NET
Thu May 28 15:38:34 EDT 2009


>>>>> "cs" == Charles Sprickman <spork at bway.net> writes:

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

you can replace all eight drives with 1TB and have a pool four times
larger.  but until you replace the eighth drive, you get nothing.  and
you must have eight drives, forever.

Another approach would be to make the 8 drives into two raidz2 stripes
of 4 disks each.  Then you could replace 4 drives, and that stripe
becomes larger, getting you a larger pool.  The problem is, with two
raidz2 stripes you get only 1TB out of your eight drives.  With all
eight disks in one raidz2 stripe you get 1.5TB.

(you should be using dual parity.)

    cs> if the box is idle and pulling like 40W the PS is not pulling
    cs> 100W out of the wall?

well that's not 80%.

the 80plus logo means more than a claim of 80% efficiency because the
80% efficient claim is surely at the power supply's most efficient
point, which will be at close to max power, while 80plus requires the
supply to be more efficient at all points along a proscribed curve of
fraction-of-capacity vs. efficiency.  just read about it.  It's not
complicated and addresses just what you want.

    cs> a "new" power DC power standard for datacenters that would run
    cs> either 12V or 5V to cabinets

that's stupid.  A good standard would be 300VDC to cabinets, 252V
batteries and breakers in each cabinet, and dual 12V switching
supplies in each cabinet.  I'd probably wire one supply to the
batteries, and the second straight to the 300V input.  This way the
battery shelf could be removed for maintenance with minimal chance of
fuckup.  Each piece of equipment should take dual 12V inputs.  

For peecees this means eliminating the powersupply entirely, not
swapping it for a 12V supply.  motherboards that eat 12V period.  Most
chips now run off 1.4V or 1.8V or 2.5V these days, especially
power-hungry stuff, so mandating that motherboards switch 12V down to
whatever voltage they need will add only a tiny cost because they are
already doing this for most of the watts they consume.  The lower
voltages in ATX are just a waste of copper.  The power inputs can go
where the PS/2 keyboard used to plug in.  cases can include 12V downto
5/3.3V switchers on their SATA backplanes to accomodate the
unfortunate SATA power standard.

Another clever design might be to wire peecees in _series_.  THese
would have some kidn of hot-plug connector, and when you remove a
peecee, the cabinet shorts its power input.  Design a power supply to
go in each peecee that cooperates with its neighbors to settle on the
minimum satisfactory amount of current passing through the whole
stack.  If one of the peecees in the stack is using more power than
the others, then all the others have to shunt power through a fat
resistor to keep current flwoing through the stack.  You can make them
use the resistor very seldom by making the current negotiation rule
rather complicated---for example, allow that each peecee can operate
with 6 - 24V across it, so if your neighbor needs more power you allow
an increase stack current flowing through yourself but decrease your
own resistance so you've fewer volts across yourself.  Once you get
down to 6V you start blowing on the resistor.  With a rack filled with
identical equipment, 6 - 24V might be enough to cover the difference
between idle and peak load.  The series stack of peecees doesn't need
the fat per-cabinet 12V supplies, so you save the cost of buying them,
as well as their minimal heat loss (though high quality DC switchers
are like >95%).  The idea is it might be cheaper to make something
that is auto-switch-itself-off tolerant of 300V than something which
can actually eat 300V, and that these imaginary datacenter standards
all presume racks full of identical equipment without fully exploiting
the similarity.  It's also funny to wire cabinets so that if one bulb
pops the whole strand goes out.

-- 
READ CAREFULLY. By reading this fortune, you agree, on behalf of your employer,
to release me from all obligations and waivers arising from any and all
NON-NEGOTIATED  agreements, licenses, terms-of-service, shrinkwrap, clickwrap,
browsewrap, confidentiality, non-disclosure, non-compete and acceptable use
policies ("BOGUS AGREEMENTS") that I have entered into with your employer, its
partners, licensors, agents and assigns, in perpetuity, without prejudice to my
ongoing rights and privileges. You further represent that you have the
authority to release me from any BOGUS AGREEMENTS on behalf of your employer.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 304 bytes
Desc: not available
URL: <http://lists.nycbug.org/pipermail/talk/attachments/20090528/a0d4ea6e/attachment.bin>


More information about the talk mailing list