> I haven't seen one post addressing the most critical issue. They need a
> Generator and some small inexpensive UPSes. The small UPSes will keep the
> servers up until the generator kicks in.

A good standby generator is not cheap. Add to that the cost of running in conduit and maintenance -- all to keep servers running that don't need to be up 100% of the time.

BOINC is like E-Mail. If a mail server is down, it can be down for a while and mail isn't lost -- it just retries. If BOINC is down for a few hours, the BOINC client just retries.

The money for a standby generator could be spent on another nice fast database server to carry an up-to-date replica (assuming the money is available, of course).

---

> > I don't know what Berkeley thinks of eBay, but you can find them there.
>
> >Well Berkeley must think nicely of eBAY -- after all, they have
> significant >funding constraints so saving money might be very much a
> priority item .
>
> Unfortunately you are forgetting that they are part of the University of
> California. And in most state funded universitys in this country, anything
> costing more than $100 must be put out for bid, a process that usually takes
> atleast 6 months. So unless they can get a corporate or individual sponsor to
> donate the equipment, getting the needed equipment is a very hard thing to do.

There is an SU041 on eBay for just under $45, so that's in the under $100 category. There are a couple of the AP9207 expanders for about $10.

At those prices you can buy some extras.

I think I saw a 3000RM or 5000RM in one of the pictures, and I'm guessing that they don't have much more power than that going into the closet.

---

> There is an SU041 on eBay for just under $45, so that's in the under $100
> category. There are a couple of the AP9207 expanders for about $10.
>
> At those prices you can buy some extras.
>
> I think I saw a 3000RM or 5000RM in one of the pictures, and I'm guessing that
> they don't have much more power than that going into the closet.
>
I saw a post yesterday that said that they have plenty of upses for the "closet", just the server that had the database on it was not in the "closet". They also don't seem to have a program that gracefully shuts down the computers when the upses do come on line. Don't even know if they make one of those?

---

Right -- a 5000VA RM probably is a bit thin though (admittedly just guessing here) -- that's why I posted a link to a 7500. But we are definitely in the same ballpark.

Barry

---

By the way, observing the back and forth about this between some folks, I caught the combination of critique and defensiveness on both sides and figured that while there is some validity on both sides -- I think the overall context of discussions like this (when nearly all of us are pretty much removed from the action onsite), ought to be along the lines of a brainstorming discussion of what happened and what approaches could be in the universe of possible means to avoid repeats.

That is, not as a criticism of what happened, nor as a rejection of discussion about what happened.

When folks get into online back and forth attack games, things can get ugly.

Been there done that -- as an online volunteer sysop for over 17 years.

Barry Schnur
Ex TeamB (Borland)
Ex Corel Volunteer sysop
Current Novell Volunteer sysop

---

> I saw a post yesterday that said that they have plenty of upses for the
> "closet", just the server that had the database on it was not in the "closet".
> They also don't seem to have a program that gracefully shuts down the
> computers when the upses do come on line. Don't even know if they make one of
> those?

I'm not a UNIX type, but there is a pretty nice piece of software called APCUPSD that handles all of this kind of stuff over a LAN -- you hook the UPS via serial cable to one of the servers, and the rest is done by UDP, if I remember correctly.

Either way, there is much software, and I don't think software was the problem.

---

> The switch has two power cords and a set of outlets. You plug each cord into
> a UPS, plug the load into the switch, and you're set.
>
> I can unplug a UPS and time it so I know how long it takes to run flat: when
> the UPS dies the load switches to the other one (which is still plugged in)
> and I can test until the batteries are flat and still have half of my battery
> capacity (in the other UPS).

For what it's worth, I plugged an analog clock into UPS #1, made sure it was carrying the load, and jerked the plug. It ran 40 minutes.

I learned about doing this the hard way.

---

> For what it's worth, I plugged an analog clock into UPS #1, made sure it was
> carrying the load, and jerked the plug. It ran 40 minutes.
>
> I learned about doing this the hard way.
>
Yes, Load testing is important. Might I offer some thoughts on your method?

1) do you know the load in Watts that the clock absorbs?

You're puter draw roughly 100 watts (give or take depending upon cards, CPU load, etc).

You might get a more realistic test of the amount of time the UPS will last if you closely match the computer load, by using a Lamp with a 100 watt light bulb.

Just a thought.

tony

---

> 1) do you know the load in Watts that the clock absorbs?
>
> You're puter draw roughly 100 watts (give or take depending upon cards, CPU
> load, etc).
>
> You might get a more realistic test of the amount of time the UPS will last if
> you closely match the computer load, by using a Lamp with a 100 watt light
> bulb.

Tony, I appreciate the comments.

The clock draws one watt by actual measurement.

My load is a total of seven computers (router/firewall, mail server, 4 web servers, and a supervisor that keeps an eye on things), plus a couple of hubs, one monitor (which is almost always off), four modems, and a couple of incidental bits. It's a total of 550w, also measured with a reasonably good meter.

I'm actually testing with my actual load, so "matching" isn't a problem.

I have two UPSes and an automatic transfer switch: I can turn off either UPS and the whole load just switches to the other one. It also means I can put the real load on one UPS, let it run to "flameout" and still have one UPS with a full battery charge.

---

> The clock draws one watt by actual measurement.
>
I didn't understand. I thought you had just one clock as load, and with just that one little bitty load, you're UPS lasted only 40 minutes. I was starting to think new batteries were needed.

have a good one.

tony

> > The clock draws one watt by actual measurement.
> >
> I didn't understand. I thought you had just one clock as load, and with just
> that one little bitty load, you're UPS lasted only 40 minutes. I was starting
> to think new batteries were needed.

Ah, yes, 40 minutes on a one-watt load would be a sign of trouble.

I've seen that happen with old batteries....

When we start swapping out servers (not sure when, but we're due for some new hardware) most of them will be replaced by Via C3 based machines. The complete server will draw about 40w.

---