There are some aspects of computer hardware and software that are unglamorous, unexciting, and unappreciated. One of the most ignored aspects of hardware on most systems is the electrical supply. At least it’s ignored until a lightning strike wipes out the motherboard, fries a modem, or power cuts off during a disk write, causing loss of data. When any of these events happen the general response is "why didn’t I get a UPS?" The truth is that Uninterruptible Power Supplies are mundane looking things that sit under a desk or mounted in a rack, seldom raising any interest until they are required. The cost of UPSs is not high, yet many people don’t bother with them, instead assuming the "it won’t happen to me" attitude. Unfortunately it does happen.
The North American power system is one of the best in the world, yet voltage spikes and brownouts are common. Power failures do happen, sometimes often. Depending on your location an average of a failure a month can be experienced. All of these power line antics can easily damage sensitive electronic equipment, causing anything from filesystem damage to catastrophic failure of the machinery and sometimes secondary effects like fire.
The technology behind UPSs is constantly advancing, improving the ability of a UPS to cope with voltage changes, delivery a constant and steady filtered voltage to connected equipment, and use battery power in case of failures. In this article we look at what a UPS does and how it achieves those goals. We also look at five SCO-certified UPS models suitable for a typical server, and test them under extreme conditions.
Types of UPSs
Most people think there are only two kinds of UPSs: standby (which runs off battery power only when power fails) and on-line (which always runs through the battery). This is a simplification as there are six or seven different topologies used in UPS design.
The most common type of UPS is the standby UPS, usually used for small home machines. A transfer switch chooses between filtered AC power or the battery, switching to the battery when the AC power fails completely. The battery and inverter circuitry (necessary to provide AC) work only when the power has failed, hence the reason the system is called "standby". The speed at which the transfer from external AC to internal battery is important, as it must be fast enough to prevent problems for any connected equipment. Standby UPSs are popular primarily because they are inexpensive to manufacture and hence sell, they are usually small, and they are efficient. Maintenance is rarely needed.
The Line Interactive UPS is used for larger systems such as servers. In this design, the battery and inverter are always active, feeding components connected to the AC outlets of the UPS. Incoming AC power is used to keep the battery charged. Since the inverter is always running, whether on battery or incoming AC power, the output is regulated and controlled, allowing for a consistent output with many fewer transients. When a brownout condition occurs, the battery can supplement the incoming power to boost it. Line Interactive UPS units are efficient and reliable, and tend not to be expensive although they are more costly than standby designs. Usually these units are available up to about 5kVA.
Standby on-line hybrid UPSs are used up to 10kVA and behave much like a standby UPS except there is no transfer time involved in switching to the battery because of the circuit design. This UPS takes both incoming power and battery circuit power at the same time and feeds it to a combiner. This design tends to be quite expensive and inefficient so it is not widely used.
Larger UPS units tend to employ different designs than the smaller units. The standby-ferro UPS used to be used up to 15kVA and was the most commonly used type for years. A transformer accepts direct AC power through a transfer switch, as well as from the battery through an inverter. After a power failure the transfer switch cuts out and all power is taken from the battery. Because of the use of a transfer output power is highly filtered and these units tend to have high reliability. However, they tend to be less efficient than many other designs and could be unstable under some circumstances. The cost of these units is high and efficiency is low, making them not too popular.
Above 10kVA to 5000kVA the double conversion UPS is almost a standard. The circuit looks like a standby UPS’ except that the inverter is used as the primary power supply and not the incoming circuit. You might think this would result in no transfer time in case of power failure but in fact these units do exhibit delays due to the inverter’s design. The double conversion design allows filtering and stable output for components. Because of the design and high power ratings, though, this design tends to run hot and therefore has reduced efficiency. In addition, this design tends to cause wear on parts and hence has a much lower lifetime than other designs.
The delta conversion UPS is a relatively new design for higher capacity units. It is similar to the double conversion design in that it has an inverter supplying the output voltage but a delta converter circuit also contributes to the output. The advantage to using a delta converter is that efficiency is much higher and power quality is better. This design is patented and therefore only available from a limited number of vendors. The delta converter can be used up to 5000kVA although the design is impractical for under 5kVA units. On a per-kVA basis the delta conversion and double conversion designs tend to be more expensive than line interactive designs.
Choosing a UPS
You’ll see lots of numbers floating around when you try to find a UPS to suit your needs. One important term you need to know is power factor (see "Terminology" sidebar). To determine the size of UPS you need, you have to add up the draw of all the equipment that will be powered by the UPS. Do not use the power or wattage rating of anything you are adding up. Instead, use the VA rating. (Power and Watt measurements do not include any reactive or harmonic effects.)
When you add up the amount of output you need from a UPS for your system, make sure the VA rating of the UPS is higher than the VA requirement. This may seem obvious but some people work on a "these both won’t be on at the same time" principle, and end up overloading the UPS. Count the total load of all connected equipment, assuming it is all powered on at the same time. Also factor in any planned purchases: external devices all need power that must be drawn from the UPS, not an internal power supply in a server or workstation.
Once you have determined the total amount of UPS you need, it is recommended practice to move up one level in power. This provides a safety cushion for the UPS itself, as well as for any future changes you may make. For example, if you were going to buy a 600VA unit, move up to a 1kVA instead.
A common gotcha in UPS purchasing is to confuse Watts and VA, as well as failing to take into account the power factor. For example, suppose you have a file server that is rated at 500W and 500VA. A 600VA UPS will handle the 500VA rating without problem. However, the power factor of .6 (see "Watt Rating" in the Terminology sidebar) results in the UPS being rated for only 360W (600*0.6), so the UPS cannot power the server.
Some UPS manufacturers have the nasty habit of not including both VA and Watt ratings on their UPS packages or literature specifications. Instead, they include only a Watt rating, which leads most purchasers to buy too small a unit because of the power factor. If the VA rating is not clear, it can be difficult for you to determine what power it can actually supply. The best advice is to avoid any unit that doesn’t clearly state a VA rating. Just as bad are ratings in VA only, without a Watt rating. Again, the best advice is to steer clear and find a unit with both ratings. Alternatively, you can usually calculate Watts by assuming a power factor of .6, which means the Watts will be 60% of the VA rating for the UPS.
Remember not to plug laser printers into a UPS: the draw a typical laser printer far exceeds the surge factor of a UPS and can cause failure or damage of the circuitry. Surge protectors are usually all that is required for a laser printer, although there are some UPSs designed with a high enough rating to handle laser printers. Inkjet printers tend not to be as bad as laser printers, but the best advice is still to leave the printer on a surge protection circuit instead of the battery-powered circuits. Some UPSs have identified outlets on the back just for surge suppression, and these can be used for printers.
How long should your UPS last? A typical unit in the 1kVA range should easily provide five years service, although the service life is reduced if the battery circuit has to kick in frequently. Both APC and TrippLite suggest a three to five year service life before battery replacement. All the UPSs we tested (and most others, except for the smallest ratings) offer user-replaceable batteries. The battery cost is the majority of the price of a UPS, though, so when time comes to replace the battery it is often more beneficial to replace the entire UPS and take advantage of developments in both UPS hardware and software.
SCO Certification for UPSs
Why would you care about SCO certification for a UPS? If all you want is a filtered backup power supply, then certification has no real meaning other than to verify manufacturer’s claims. Most UPSs you will want to use have serial ports that allow you to connect the UPS to your machine. Manufacturers provide software that monitors the UPS all the time and can send alarms to your terminal when power conditions change. Alerts and other warnings can be broadcast in case of power failures or battery problems, for example. The better software packages can shut the server down when conditions warrant, too, saving your machine from an ungraceful crash.
SCO certification for the software has definite value. It means the software has been tested under SCO UnixWare 7 and SCO OpenServer 5 and that the UPS can properly control the server. Also, status messages and alert levels can be set to work properly with the operating system. Of the UPSs we tested, all but the MGE unit were previously SCO certified. However, we downloaded the complete set of compatibility and certification tests from the SCO web site (http://www.sco.com/hbk) and conducted the full suite test on all UPSs, including MGE’s. All passed perfectly.
We decided to test a typical server-supporting UPS. The most common rating for this setup is in the 1kVA to 1.5kVA range, which allows powering of a well-equipped server with auxiliary equipment and network switches. We requested UPSs in this range from eight vendors. Three vendors (APC, MGE and TrippLite supplied units immediately). Two others (Best and PowerWare) arrived just before deadline. The remaining three refused to send units for testing. Draw your own conclusions!
The UPS market is dominated by two companies: APC and Tripp-Lite. Sales figures are difficult to obtain because Tripp-Lite is a privately held company. However, APC outsells Tripp-Lite partly because of APC’s immense penetration into the market. The remaining companies are quite a bit behind these two leaders. Best is third, although their sales are about a fifth of the two big companies. MGE is a relative newcomer to North American, arriving from France and trying to create a presence in this market. Best and PowerWare both market to the smaller UPS niche, selling units in the 5kVA or smaller range. APC and TrippLite, on the other hand, cover a larger range.
We used two servers for this test. The first machine was an ALR Revolution 2XL equipped with dual Pentium II 266Mhz CPUs, 128 MB RAM, three 9.1GB hard drives, a CD-ROM, and HP SureStore 5000 internal DAT. External devices include a NEC 6-CD changer, HP SureStore CD-Writer, Iomega Jaz drive, and an HP ScanJet 4. Also on the same circuit was an HP Ethernet switch, ViewSonic P815 21-inch monitor, and a bank of six 56k modems. The operating system was SCO OpenServer 5. The second machine was a custom-built ASUS motherboard with dual Pentium III 500MHz CPUs, 256MB RAM, eight 9.1GB SCSI-2 hard drives, Sony AIT tape drive, CD-ROM, and internal Jaz drive. External devices included another HP Ethernet switch, TTI’s Series 2300 dual-DLT backup subsystem, and ViewSonic P810 21-inch monitor. The operating system was SCO UnixWare 7. Both sets of electrical components were connected to the UPS under test, one at a time. When the vendor supplied SCO drivers for their UPS interface, they were installed and tested.
The test suite started simply enough: we used the UPS in routine daily use for a few days, monitoring the output power supply through recording equipment connected in series with the server and components. Failing the incoming power (by shutting off the circuit breaker) allowed us to test the battery life under full and partial loads, as well as to evaluate how well the monitoring software reported conditions and eventually shut down the server. To properly test the limits of the UPSs we assembled a suite of equipment from a protection engineer’s lab. We could vary the incoming voltage and current infinitely over a range from 0 to 400W and 50,000V. Since the equipment allowed repeatable tests to be set up, we created a test suite that started duplicating brownouts and overvoltage conditions starting mildly and growing to extremes that would be encountered in a typical household or business power supply. We exceeded these values by 50% to continue the tests, being careful not to damage the UPSs. Finally, we fed a series of spikes of varying intensity and duration to the UPS to check its filtering abilities, recording the output during each test.
The result of the tests was encouraging. Each UPS we tested could handle all the voltage and amperage changes we threw at them, even those well outside the range you could expect to experience at your wall socket. They all performed admirably on the spikes as well, primarily because of the design of the UPSs which provide battery-inverter power rather than direct wall power. (All the units seem to operate in line interactive mode. Standby units tend to be much smaller than our requested sizes and would have behaved notably different.) With each unit having passed the SCO certification tests as well as our lab bench tests, we then moved on to evaluate the software and UPS designs themselves. These are the real deciding factors for most users when they look for a UPS. The results of our assessments are given below.
APC is the best-known name in UPSs for a reason: they make attractive, reliable units which are available in practically every computer store in the country. They got to be everywhere through good marketing. The SmartUPS line became popular because it does what it is supposed to: protect your equipment.
The SmartUPS is an attractive unit. It is the traditional long, low, thin UPS shape but nicely rounded with an attractive front panel. On the front are two vertical bars of five lights showing the battery capacity and current draw, as well as six status lights. Two switches turn the unit on and off. The back of the unit has six sockets, all closely spaced, so wall-warts have to go on a separate power bar plugged into the UPS. The back also contains a "smartslot" which allows accessories to be plugged in. A supplied cable can be used to add external battery packs to the UPS, too.
The APC UPS ships with PowerChute. The CD-ROM includes software for numerous platforms including SCO Unix. Documentation is included on the CD as ASCII and .pdf files. The Windows versions of PowerChute are much better than the SCO version, offering integration with many NT tools like the Performance Monitor. Unattended shutdowns and reboots can be scheduled with PowerChute, and front-panel status indicators are visible on-screen through a dialog. The latest version of PowerChute and an expansion card that plugs into the smartslot add SMTP notification through e-mail, which will help administrators of large sites.
When failed with a full load of 1400VA the battery lasted fourteen minutes. With only the server and a few external devices powered (total load of 600VA) we obtained forty-one minutes from the battery. With just the server and a network switch operational (manually powering down other devices to save battery power) we managed eighty minutes.
If there’s a weak point in APC it is the company’s technical support. Just prior to starting this review one of our fourteen-month old APC UPSs started failing at intervals, providing no battery backup after one particularly nasty shutdown following an electrical storm. Front panel light diagnostics indicated a battery problem so we contacted the technical support number for replacement information. After an interminable amount of time on hold we were told we needed a new battery but we couldn’t order a replacement from that number. Further, when asked why it failed after just over a year on the job we got brushed off and told nothing under warranty covered the replacement.
APCs SmartUPS is a well-known product in the market and our review unit performed excellently, outlasting our Top of the World winner from TrippLite. The relatively poor SCO implementation of the PowerChute software prevented the SmartUPS from making it to the top of the list, but these UPSs are definitely close contenders, especially under Windows.
The Best Fortress UPS arrived late and as such spent only a short time in the testing labs. Because of Best’s popularity, though, we did want to include it in this review. The latest incarnation of the Fortress product line looks very similar to its earlier versions. The unit is nine inches high, seven inches wide, and eighteen inches long, designed to sit under a desk or next to a computer case. (Many owners of this type of case, such as those manufactured by Best and APC, find that not only are the UPSs excellent protection for their computers but they also make great footrests!)
The front of the beige case has two pushbuttons and a vertical stack of lights. (Older Best power units had digit displays that continually showed accurate numbers instead of the approximation a stack of LEDs offers, but presumably that style of interface costs too much for most consumer units.) The back of the Fortress has only four outlets in a staggered arrangement, although one outlet can handle wall-wart power supplies. As usual, power bars are necessary to plug in equipment. The back also has a DB9 serial port connector as well as a slot for an optional SNMP module (our unit came with one). Installing the SNMP module was simple and took two minutes.
Best has offered their CheckUPS software with their UPSs for years, and CheckUPS II v 3 that accompanied our unit was noticeably better than older versions. The graphics are excellent, and support is provided for several operating systems on the same disk. When the SNMP interface is installed, a network manager can easily obtain information about any equipped UPS on the network. Automating shutdowns and alarms was easily than with the majority of software we tested.
When failed with a full load of 1400VA the battery lasted nine minutes. With only the server and a few external devices powered (total load of 600VA) we obtained thirty minutes from the battery. With just the server and a network switch operational (manually powering down other devices to save battery power) we managed seventy-five minutes. The latter number was the highest of all the tested units (backing up Best’s prominent claim on the shipping carton "Now with Free longer runtimes".
The Best unit came closest to giving the TrippLite competition for "Top of the World" rating, missing only because of the second serial port on the TrippLite. If that’s not important to you, consider the Best unit carefully. It provided excellent runtimes, good software (especially under Windows) and the box is attractive to look at, too.
MGE is a newcomer to the North American UPS market but they are well established in Europe. The Pulsar UPS we were provided with is a rack-mount single-height unit, with an attractive curved front panel. For the rating, the unit is surprisingly slim. Physically, the UPS is three and a half inches tall, standard rack-mount nineteen inches wide, and nineteen inches deep. The front panel has a row of six lights for different status conditions as well as a front on-off button. A second function button allows you to switch of the audio tones when activated, as well as measure the load percentage using the front lights as a scale.
The back panel of the Pulsar has four outlets, closely spaced, as well as a fuse socket, DB9 serial port, and a fan outlet. A connector allowed the Pulsar to be attached to addition battery subsystems for extended life. Rack rails are included in a separate package so you can leave the Pulsar as a floor or desk unit if desired (bearing in mind circulation and heat requirements). There is room on the back of the Pulsar for an SNMP card, although our review unit didn’t have the module installed.
A CD-ROM with MGE’s Solution-Pac software accompanies the UPS. As you would expect, software is available for all the Windows platforms as well as most UNIX workstations, including SCO Unix and SCO OpenServer. The software is pretty good but isn’t in the same league as TrippLite’s PowerAlert, for example.
The Pulsar UPS performed well in our duration tests. When failed with a full load of 1000VA the battery lasted sixteen minutes. With only the server and a few external devices powered (total load of 600VA) we obtained forty-five minutes from the battery. With just the server and a network switch operational (manually powering down other devices to save battery power) we managed eighty-three minutes.
The MGE unit was easy to work with, actually attractive in our rack, and worked well in our tests. We hope MGE manages to penetrate the market so that these products will become available to many more administrators and users.
Powerware 9 Prestige
You may not be familiar with the name Powerware, but you’ve most likely heard of Exide, Powerware’s previous name. The Powerware 9 Prestige UPS series all have the same look. The review unit was a rectangular box measuring about ten inches tall, six inches wide and seventeen inches deep. It’s a beige box that doesn’t manage to look utilitarian thanks to a well-designed case. The case can be placed vertically as a single unit, or it can be stacked horizontally one on top of another with other Prestige line UPSs, making a nice looking pile.
The front of the UPS has a blue face with controls and light in the top right corner. Two contact switches control power on and off, and a set of stacked LEDs act as battery status indicators. The back of the unit has only four outlets closely spaced, with no room for thick plugs. Power bars are almost a necessity with this design. The back also sports a DB25 serial port (the only UPS we’ve ever seen with a 25-ping port) and a connector for attaching auxiliary battery packs. The design is curiously effect and attractive, although we really would have liked more sockets on the back panel, especially for a unit with this rating.
The Prestige package includes a serial cable (DB25 to DB9 only), power cord, and an "Operator’s Manual". For some curious reason the majority of the manual is in English except for the safety warnings which are in eleven different languages on as many pages. Connecting the test unit to the server required two power bars to handle all the plugs.
The software that accompanies the Prestige is FailSafe III which provides drivers for a wide variety of operating systems on a single CD-ROM, a very nice touch. The software is graphically based, attractive and easy to work with. It shows on-screen status reports from the UPS which also include a block diagram of the internal structure, showing current routing. The on-line documentation claims FailSafe III can be used with UPSs from other vendors but we couldn’t get the software configured to handle any of the vendors we tested in this review. However, we only tried for a short while, as the unit was late in arriving.
When failed with a full load of 1400VA the battery lasted seven minutes. With only the server and a few external devices powered (total load of 600VA) we obtained twenty-one minutes from the battery. With just the server and a network switch operational (manually powering down other devices to save battery power) we managed forty-two minutes. These results were lowest of the tested UPSs.
PowerWare’s brochures talk of many add-ons and extras for the new line of UPSs that would make this an interesting choice for roll-over over a network. We focused on the single test unit, though, and found it to work well. The software was impressive, the UPS attractive. Only the poor test results marred the overall impression of this UPS.
TrippLite SmartPro Net 1400
TrippLite’s SmartPro won our Top of the World award for two reasons: two serial ports that allow two different machines to be connected to the UPS, and the software’s ability to remotely manage UPSs. It also won our "least graceful" award for the boxiest, most inelegant design!
The SmartPro 1400 is designed to stand vertically as a small tower. The unit is about sixteen inches tall, six inches wide, and eleven inches deep: a small footprint considering cthe power rating. The beige box has a set of five status lights on the front, halfway up the face, along with two slide switches. The back panel has eight sockets tightly arranged in the top half, and two DB9 serial ports below them. The sockets are placed tightly together leaving little room for larger wall-wart power supplies. We used a supplemental set of power bars to plug in our modems. The rest of the case looks like a typical industrial box: exposed screws on the battery compartment (lower front) and ventilation slots on the top. There’s no attempt to be graceful or elegant, which reflects the engineering focus of the company.
The two serial ports allow you to connect the UPS to two different machines, a feature we initially didn’t appreciate much until we realized that this allows two smaller machines to be powered (and managed) by a single UPS. TrippLite is the only company to provide this capability and we hope it catches on. TrippLite also seems to be the only company to provide single-outlet control through their PowerAlert software, allowing you to shut down some sockets and keep others active to shed loads in case of a problem.
The PowerAlert software included free with the UPS includes support for practically every UPS on the market, including those from TrippLite’s competitors. If the UPS is networked, PowerAlert can mange it. With our network of thirty PCs, all protected by UPSs, we could gather data for the entire network into one report. Apart from its ability to manage any networked UPS, the shutdown routines were fully customizable. A remote reboot feature allows a system administrator to reset locked-up hardware over a network, which is a wonderful feature. The software is obviously focused at the Windows markets but the SCO drivers worked perfectly.
The front panel of the UPS has only five lights but they suffice to show over a dozen different conditions. A self-test button lets you check the battery and UPS performance at any time even with a load attached. The battery can be hot-swapped, which also is useful.
When failed with a full load of 1400VA the battery lasted ten minutes. With only the server and a few external devices powered (total load of 600VA) we obtained thirty minutes from the battery. With just the server and a network switch operational (manually powering down other devices to save battery power) we managed 65 minutes.
With PowerAlert’s individual outlet control and network reporting, as well as the SmartPro Net 1400’s dual machine connections, we had no qualms about awarding this UPS the Top of the World rating. We’re buying the review unit.
All the UPSs we tested worked perfectly on our failure, spike, and over-voltage tests. They all kicked in smoothly and easily to handle under-voltage conditions, as well. From the point of view of functionality (how they protect your equipment) there’s not a lot of difference we found in our testing. The software that accompanies all the UPSs has gotten much better over the years, too. We would be happy with any of these UPSs and software libraries. You can easily choose on the best value (most inexpensive, in other words) and not be disappointed.
Choosing a Top of the World was tough because we had to rely on qualitiative aspects for the most part. If we went purely by appearance, our Top of the World winner would have been dead last (the TrippLite unit looks very utilitarian; maybe that’s the effect they’re after). However, the TrippLite software was just a little better, the second serial port on the back just a little handier to have, and the test results almost as good as the Best to gather first place. The Best Fortress comes in very close in second (it would have been first based on test results alone) and the SNMP module is an excellent add-in for network managers. That doesn’t mean the other three units fail in any way, though. They all are excellent products, and all deserve attention.
Terminology: Learn your V, A, and Vas
Power Factor: in AC circuits some reactive or harmonic current is involved which gives a higher apparent power than simply multiplying volts times amps for power. The difference between apparent power (V * A) and actual power leads to the power factor. Apparent power is given as a VA rating, which must be multiplied by the power factor to get the actual power. This effect can be considerable: a PC magazine test found apparent power 50% higher than actual power.
Surge factor: the momentary overload capacity of a UPS. This measures the ability of the UPS to handle short overload conditions, usually involving startup voltage demands by equipment. Most disk drives, for example, have a high surge factor requirement as does anything with a motor. Typically a surge factor is about 50% higher than the steady-state rating, so a 1kVA UPS should be able to handle 1.5kVA for a very short amount of time.
VA: volt-amps are the apparent power used by equipment, as opposed to the actual power (Watts). The difference is because of the power factor.
Watt: unit of energy. Watt ratings determine the actual power.
Watt rating: most computers use a capacitor-switching power supply which has a power factor of about .60, meaning the Watts drawn by the computer are about 60% of the VA rating.