|
|
| Author |
Message |
janc
|
|
UPS
| 
Jan 5 00:16 UTC 1999 |
Right now Grex's computers are plugged into a thing called a "power
conditioner". It's a heavy-duty gadget that filters just about any
imaginable power spikes. It does a nice job of protecting our
equipment, but I understand it also wastes a fair amount of power. The
theory is that if we replaced it with a modern UPS, then we'd still get
good protection, with much less wasted electricity. Depending on how
much power it saves, it may pay for itself in a few years. Plus, of
course, you get the benefit that the UPS, unlike the power condition,
will keep Grex running through short power outages.
I wouldn't support buying the UPS just to keep Grex running in power
outages. We don't have all that many power outages. But saving on
electric bills would be a real benefit.
We need to consider if we want to do this. Obviously we need some
numbers first.
|
| 83 responses total. |
steve
|
|
response 1 of 83:
|
Jan 5 01:26 UTC 1999 |
That power conditioner is an excellent heater in winter; there
were many times that I put my feet on it when Grex lived in the
Dungeon, during the winter.
In the summer it looked more like a device from the devil, as
it radiates a *lot* of heat.
Ferro-ressonant conditioners have many benefits, but power
conservation is not one of them. I cannot imagine that we are
wasting less than 150 watts on this, compared to a real UPS.
I think electricty is 10.5cents per kWH. Given that there are
8.760 hours in a year, a one watt drain on Grex costs us about
$0.92. We can use that as a good starting point. This means
that a 100w savings amounts to $92. If we asusme just a 100w
savings, within three years we'd see a savings, but I think its
closer to 150 - 200w that our conditioner is eating, so the
financial payback is that much closer.
The other payback is harder to measure, but it includes
- less heat whever we are,
- increased hardware reliability by lessened temperature,
- increased comfort for staff in the summer (with the exception
of marcus who probably likes it),
- increased protection during real problems such as half second
outages that the conditioner can't handle, and allows for a
measure of protection against other strange electrical gifts
the power company may give us.
We need to get a power meter again and look at the conditioner.
|
jared
|
|
response 2 of 83:
|
Jan 5 03:01 UTC 1999 |
a APC 2000 is $800 or so last I checked. I don't know if this
would work for grex (beast w/ way too many fans, that sucks power
excessiveley), or for how long, but I figure I can keep my stuff
alive here for about 4 hours.
I would recommend something like this for the grex stuff, it
would serve the power conditioner issue, along with enough to
get grex through most outages such that the disks won't just
totally die.
|
steve
|
|
response 3 of 83:
|
Jan 5 03:04 UTC 1999 |
That sounds about right to me.
|
i
|
|
response 4 of 83:
|
Jan 5 03:27 UTC 1999 |
I have some experience with APC as a dealer, and am not very favorably
impressed with them. But brand and size are messy parts of a serious
UPS decision, and we're a ways from having to make those calls.
|
steve
|
|
response 5 of 83:
|
Jan 5 04:58 UTC 1999 |
Thats good to know, if you've had not so good expereinces with
them. Somewhere on the web I saw a comparison of UPS's from an EE,
whose sole decision for what one to use was based on the output of
a scope. I think I might remember here that is. It would be a good
place to start.
|
i
|
|
response 6 of 83:
|
Jan 5 06:01 UTC 1999 |
Hmmm. Do we know what we'd power off the UPS? Some stuff needs a nice,
shapely sine wave, other gadgets really don't care if they're fed a
sloppy square wave. (And some UPS's output waveform changes with the
source & load.)
(My philosophy is that we should come up with a capacity & runtime and know
what we'd need shutdown-software-wise before we get much into brand names.)
|
devnull
|
|
response 7 of 83:
|
Jan 7 06:16 UTC 1999 |
What I've seen of APC's shutdown software running on scumos 5.5 is that
it actually made the relavent sparc less relibale when it was running
than if it wasn't running; I think if the power went off and came back, it
went into an infinite loop, which drove the load average up to one if the
system wasn't otherwise occupied.
It's possible that they fixed their lossage in the last few years, of course.
|
jared
|
|
response 8 of 83:
|
Jan 8 04:44 UTC 1999 |
hmm.. comparisons about computer things that are over a "few years"
in age. sounds like time to recheck, because of that whole
technology constantly evolving thing
|
djf
|
|
response 9 of 83:
|
Jan 9 14:40 UTC 1999 |
I'd also suggest staying away from APC, if only because there are
better (and more friendly) vendors out there.
I bought a Best Power UPS last spring and have been very happy with
it. They supply source code to their UPS daemon and will provide
thorough documentation of their serial protocol free on request. I
bought a Best Fortress 1420VA, and I believe they make larger ones.
Before purchasing my Fortress I had very positive experiences with
their FerrUPS (higher-end) line when I was working for ANS. And,
FWIW, the Best Fortress is one of the most highly recommended models
in the Linux UPS-HOWTO. Check www.bestpower.com for more info on
their current models.
There's little question that a Best UPS is going more expensive than
"similar" APCs but from everything I've read and observed about them,
they are well worth it. Since it is such a clear single point of
failure, buying a marginal quality UPS would be worse than the current
situation, IMHO.
|
steve
|
|
response 10 of 83:
|
Jan 9 17:55 UTC 1999 |
Thanks for the data, David. These days in the electronics world
it seems like the brand that charges 20% more is the higher quality
brand. We'll look around before lots before buying.
|
lilmo
|
|
response 11 of 83:
|
Feb 6 20:03 UTC 1999 |
So have we considered the issue raised by #6, yet?
#4 makes a good point that brand and size are not the first things we need
to consider. We need to figure our needs, first, and outline our wants, too,
and see what brand/size meets those most economically (remembering that the
degree of reliability and service goes into the wants/needs category, to be
met before considering price).
|
steve
|
|
response 12 of 83:
|
Feb 6 22:31 UTC 1999 |
We've always been aware of the differences in the quality of the
power generated.
As for the capacity, I'd like to see if we can't find a UPS able
to handle everything. Right now we're at just over 900w and I'd
like to have extra capacity, around 1500w or so. We need to see
what the larger UPS's cost nowadays. I know the prices have come
down but I don't know if they're in the area we can afford yet.
|
scg
|
|
response 13 of 83:
|
Feb 6 23:29 UTC 1999 |
We could also go for two smaller UPSs instead of one big one.
|
steve
|
|
response 14 of 83:
|
Feb 7 00:31 UTC 1999 |
There is a real problem with that: if both are running at once
really odd problems can occur if they're out of sync with each
other, in terms of the voltate sine wave they produce. It's fine
to have two UPS's powering equipment that isn't directly connected
to each other except for an ethernet cable, but I'd very much
question the Sun-4 on one and the disks on another. This isn't
to say that we couldn't parition things in the right way between
two of them, but its something we'd always have to remember about.
|
scg
|
|
response 15 of 83:
|
Feb 7 00:52 UTC 1999 |
I wasn't suggesting putting the Sun 4 on one and the disks on the other, but
we've got a couple of computers, plus modems, terminal servers, routers, etc.
in there, and there's no reason all of those need to be on the same UPS.
|
devnull
|
|
response 16 of 83:
|
Feb 7 03:52 UTC 1999 |
I don't disagree that putting equitment on several UPSes may be a bad
idea, but I'm skeptial that there will really be an issue with putting
the sun and its disks on different UPSes and having the sine waves be out
of phase or something. These devices all have power supplies which are
supposed to be making nice smooth DC out of whatever AC input they get...
|
steve
|
|
response 17 of 83:
|
Feb 7 03:57 UTC 1999 |
I've seen two systems that were positively weird when broken up
on a couple of UPSs. Rearranging things cleared up a series of
problems which is what caused me to read about this.
|
i
|
|
response 18 of 83:
|
Feb 7 12:47 UTC 1999 |
In general, two smaller UPS's won't be cheaper than one big one (unless
you get lower-quality little ones). Unless the system has *redundant*
power supplies designed to be fed from independent power sources, putting
two UPS's on it is just multiplying the ways that things can go wrong.
A partitioning scheme like the steve's are discussing would make more
sense. Remember that two systems on one UPS means either a complex
power-failure shutdown situation or one system's gonna get it's power
cut without warning when the other turns off the UPS. Add an emergency
light or two.
Other notes: UPS's are generally rated in VA, and 1VA is about 0.7W.
Past about 2500VA, the bigger-means-cheaper-per-unit-power rule of
thumb starts breaking down.
|
devnull
|
|
response 19 of 83:
|
Feb 7 20:21 UTC 1999 |
Is there any equitment other than the sun-4 that actually wants to be
warned of the power outage so it can shut itself down?
|
dang
|
|
response 20 of 83:
|
Feb 7 22:48 UTC 1999 |
Ideally, we'd shut down *all* the computers in there, however many there
are at the time. However, I'm not sure that we want the UPS to shut
down grex, as I've heard bad things about these things on Suns, sunos in
particular. The two real reasons for UPS, as I see it, are to smooth
bad power, thus avoiding unnecessary crashes, and to keep things running
through short (10 sec to 1 min) power outages. Much more than that, and
we couldn't keep things going anyway.
|
steve
|
|
response 21 of 83:
|
Feb 7 22:51 UTC 1999 |
When we have them the kerberos and mail machines really won't
want to go down during a power interruption.
Right now, we'd want the Sun-4, terminal server, modei and Pipeline
to stay up. But in the future things are going to be different. Thats
what we need to remember and plan for, either in a larger ups now, or
making sure that several will work in the future.
|
keesan
|
|
response 22 of 83:
|
Mar 27 01:22 UTC 1999 |
By definition, Jim agrees, 1 VA = 1 W. He is laughing. Perhaps you are
taking into account rms or some sort of fluctuation.
|
scott
|
|
response 23 of 83:
|
Mar 27 12:39 UTC 1999 |
1W == 1VA falls apart on things like inductive loads. Computer power supplies
are rather reactive, so "power factor" becomes an issue.
|
rcurl
|
|
response 24 of 83:
|
Mar 27 20:14 UTC 1999 |
Scott is right. A very good capacitor or low resistance inductor could
be 'consuming' 100 VA but just 1 W or less.
|