general questions about 7.0 and computer efficiency......
m.seaman at infracaninophile.co.uk
Wed Aug 6 09:55:06 UTC 2008
Gary Kline wrote:
> I kep track on the load on my main server, and it is rarely above
> 0.20. If the load is a poor metric of power use, what is
> better? (My new `Watt-o-Meter' is checking the power right now,
> but I would like to know what drink the most juice: disk,RAM,
> processor, OpSys? Number of hit/hours? I want my upgrades to
> be as cost-effective as possible, in other words.
Generally, the faster anything runs the more power it consumes. It
takes more energy to switch on-off (ie generate a digital pulse) the
faster you try and do it. Similar considerations apply to disks --
the faster the platters spin and the faster you need to move the heads
about, the more energy it takes. Actually, for disks the physical size
of the disk has quite an effect there too: small form-factor disks as commonly used in laptops and various HP servers are more efficient GB
for GB than the equivalent 5.25" standard drives.
There are exceptions to this rule -- for instance early model Xeon
processors ran really hot, meaning they were power pigs. Current
AMD and Core2 processors use lower voltages internally and consequently
are a lot more thermodynamically efficient.
If you want to save energy, there are two pretty useful strategies
for the home user:
* use eg. laptops -- these are carefully designed to be power efficient
in order to give the maximum battery life. They also effectively
have a built-in UPS. Similarly you can use low power appliances
built around specialist chipsets like the VIA Eden range.
* take advantage of the massive computational power of modern high-end
kit, and consolidate a number of old machines as virtual hosts on
one physical server. There are several freely available host systems
you can use -- Xen is an interesting choice, but AFAIK it doesn't
support FreeBSD as the *host* -- it's fine for *guests* though.
VMWare ESX is apparently available at zero cost as well.
Unfortunately even though the running costs will be lower, the up-front
costs for either of these strategies will be higher than simply carrying
on as before. There's also a significant 'all the eggs in one basket'
problem with virtualization.
PS. If you happen to be running a whole machine room, then there's a lot
of energy efficiency to be recovered by looking carefully at power
conversion. A typical large scale UPS will run at about 85% efficiency.
An individual computer power supply will usually be less efficient than
that. So even before you've got to the 5 and 12V buses on your mother
board, 28% or more of your input electrical power has been dispersed as
Then consider how a high-end UPS works. It takes in incoming 240V 50HZ
AC power and internally converts it into 12 or 24V DC. Some of this power
goes towards charging its battery units, but the rest is /converted
back to 240V 50Hz AC/ for supply to the computer PSU, which then /converts it back to 5 and 12V DC again/. Converting back and forth so many times
is insane. Better for the UPS to provide low voltage DC electrical power
-- for historical reasons it's usual to provide 48V DC (the higher voltage
means there's less transmission loss due to resistance in cabling, plus
with this sort of set-up minimizing the length of cable runs is good practice too). Unfortunately, trying to find a commodity PCs (even
server class machines) with 48V DC PSUs is next to impossible and damnably
expensive even so.
Dr Matthew J Seaman MA, D.Phil. 7 Priory Courtyard
PGP: http://www.infracaninophile.co.uk/pgpkey Ramsgate
Kent, CT11 9PW
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 258 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20080806/34998232/signature.pgp
More information about the freebsd-questions