Server power supply smoked; router LAN/WAN routing died

[asavage]

At about 1400 PST Thursday, the server's power supply fried. I came home to the acrid smell of burnt electronics. Joni (the server for this site) had puked a cap in the power supply.



The PS' tag said 1999, and that server has been running continuously since 2002 at 100 CPU load (I ran the SETI client until a few months ago; thence I switched back to the Bovine Client). Though it's hand-picked hardware, it's still consumer-class bits with which I'd built it, so I'm OK with it dying every so often. I'd love to upgrade both the "main" boxes, but nearly everything would have to go, as Joni & Tori are both so old that they're still running PC133 RAM, 1.2Ghz Athlons. All SCSI too, in a whole pile of identical external drive cabinets. So "upgrading" them would really mean new boxes, and I'm not up for that -- yet.

At the same time as the PS failure, the cheesy ActionTec DSL router's LAN side died (10Mb/s connections it'll recognize and pass through to other local LAN connections, but nothing will traverse its internal bridge to the outside world, and no 100Mb/s connections work, which is most of my network, so it's toast). I've purchased a used NetGear router and installed it, and so far, so good.

The data seems to all be fine -- there were a couple of odd "last post" dates, but they corrected themselves after I posted/deleted a couple of junk msgs. The data is backed up nightly on another box that was unaffected (and I retain seven days worth of backups). Worst case is that Thursday's posts might be lost, and I doubt even that much. The server's data is probably fine, even without the backup.

So: new (used) power supply in Joni, the webserver; and new (used) DSL router that I hope won't require a power cycle every two days like the ActionTec did.

Oh, yeah: if anybody cares, the PS fan on my personal box is squalling too, now that I shut it off -- it was fine until then. So I have to pull it apart and lube it -- and clean out the dust. That means I'm typing this on computer number three, normally a test box only. Oh, well.

Meanwhile: expect some more outages this weekend. I have to do UPS battery maintenance which requires shutting down everything connected to the three UPSs -- which is pretty much everything. Every six months. It's time to pull the covers and fire up the vacuum too, the dust is thick after winter. So for this holiday weekend at least, expect the server to be unavailable periodically.

[philip]

Is there a rule of thumb for sizing a power supply .... with longevity in mind?

By this I mean if the nominal power requirement is 150 watts with surges to 200 watts, you would pick a power supply of what rating?

-Philip

[asavage]

For a wide range of demand, bigger is always better. There is no benefit (other than monetary) to running a SMPS (SwitchMode Power Suppy) close to capacity. A PS of a given load capacity has commeasurate cooling capacity, and when you run it well under that capacity it runs much cooler. Heat is the killer, and if you can run it cooler, it will last longer.

There is a lower limit of load vs capacity, but it's under 20% of capacity, hard to fall short of that with consumer-class power supplies. The regulation limit of the SMPSs that I've had to repair is somewhere in the neighborhood of 10%. Run it with a load less than that, and they'll either fall out of regulation or fail to start at all.

Equipment where weight is a factor usually use designs that run closer to the capacity of the unit -- laptops come to mind.

[asavage]

if the nominal power requirement is 150 watts with surges to 200 watts, you would pick a power supply of what rating?

I don't buy replacement PSs with less than 250w -- that's my absolute lower limit for anything that can physically accept a larger PS. There are some proprietary case designs where you do not have a choice, you use what the mfgr sells you, and that's usually OK because those mfgrs that don't use the "clone" footprint are generally selling things that are robust -- server-class units. Compaq, IBM, and HP all sell both consumer-class (Presario, Aptiva, Pavilion, respectively) and business-class (Prosignia, Deskpro etc.; NetFinity, Intellistaion etc.; Proliant, Vectra etc.) PC hardware. There is a huge difference in both design and quality between the two lines. There's a good reason why those mfgrs don't use the "clone" form factor and specs for their business-class hardware, and I agree with it: sole-source for hardware replacements means the stuff costs three times "clone" parts, but works right 100% of the time, vastly reducing further support costs. BTDT, for several years.

To answer your Q, I generally aim for a 300W supply for generic use. This has worked out fine for me, because even though I use old, power-hungry Athlon CPUs (which, IIRC, draw almost exactly double the current of an equivalent Mhz Intel, but which also outperform Intel very well), I have the most power-hungry peripherals -- the HDDs -- in external enclosures with their own PSs.

Bear in mind that I got out of the PC business several years ago, and am quite out of it when it comes to hardware built after 2002, so my advice on the topic should be considered with that in mind. For example, P4 CPUs have required a 400W PS for years -- per Intel's design target spec.

I was more concerned with the old ATX standby supply current, which a lot of clone PS mfgrs couldn't meet. So many ma on the standby 5v line. Had quite a few boxes that would not start up because there wasn't enough standby current available. But that's all old tech now.

[sidebar: I ran all my hobby websites on a single IBM PS/2 Model 90 (8590) for a couple of years, back in the '97-2001 era when DSL was new. A single P90 (expensive upgrade from the "stock" P60/66 that the 90 shipped with) would provide quite decent performance, and the PS/2 line was very heavily built and easy to work on. I still have two gutted Model 80 tower cases in storage, with a board across them, making a nice shelf unit. The Model 8580 tower cases had a built-in fold-down carry metal handle, with a sticker next to it that reads "warning: 42kg" or something like that. Fully populated, they were very heavy. They maxed out at 486-33 however, and even extensive aftermarket mods would only do the clock-triple thing, so I never used a Model 80 on my server "cluster", but they sure made a nice workstation!

The triple-stack of Model 90s (webserver, ftp server, backup domain server) got retired and replaced with this single clone in 2001. The reduction in noise alone made it worthwhile. The clone itself got upgraded to 1.2Ghz two years ago, which is about five times the HP it needs for what it does.]

The new PS is a "premium" used unit by PC Cooling, a brand recognized for mfgring higher-end "clone" PC PSs. It does run cooler than the old Taiwanese unit, but it's also imparting a lot more vibration into the shelf upon which the case resides. This is important because all this hardware is in the room where I sleep! I just had to rob one of my foam gardening "kneeling pads" from the van to put under Joni to eliminate the transmitted vibration. You could only hear it in a quiet room, but my room is . . . quiet.. The foam shut it up. Now I have to go buy a fifth kneeling pad -- one at work, one in van, one under Joni, one under the electric space heater (rattled the floor when the fan came on) and one somewhere where I can't find it anymore.

[philip]

if the nominal power requirement is 150 watts with surges to 200 watts, you would pick a power supply of what rating?

I don't buy replacement PSs with less than 250w -- that's my absolute lower limit for anything that can physically accept a larger PS. SNIP

To answer your Q, I generally aim for a 300W supply for generic use. This has worked out fine for me, because even though I use old, power-hungry Athlon CPUs (which, IIRC, draw almost exactly double the current of an equivalent Mhz Intel, but which also outperform Intel very well), I have the most power-hungry peripherals -- the HDDs -- in external enclosures with their own PSs.

Bear in mind that I got out of the PC business several years ago, and am quite out of it when it comes to hardware built after 2002, so my advice on the topic should be considered with that in mind. For example, P4 CPUs have required a 400W PS for years -- per Intel's design target spec. SNIP

Ok ... how do you go about determining the maximum demand placed on the power supply? The reason behind my question is my power supply's cooling fan speed drops (audibly) during those moments when I ask the processor to do some heavy lifting. Yes, the motherboard has temperature sensitive fan speed governance. At any time, the fan reports to be spinning from 2,200 rpm to has high at 5,500 rpm. At the high end, the fan whistle becomes a scream.

The CPU is an Itel P4 Hyperthread 3.0 gig.


-Philip

[asavage]

I can't help with this one, someone else will have to jump in here.

[asavage]

The NetGear DG632 that I bought lasted almost one week. It was locked up this morning (Sat) and power cycling it would only last about ten minutes. Traded it in for a NetGear DG834. Fingers are crossed.