Low-power computing: a tech guide
This project came about at the suggestion of a charity, Computer Aid International, which collects unwanted PCs from UK businesses, refurbishes them, and then ships them for reuse in education, health and not-for-profit organisations in developing countries. Unfortunately, traditional desktop PCs, as used in western companies, have relatively large power requirements — especially when paired with old-style CRT monitors. At the same time, power supply in developing countries is often unreliable, particularly in rural areas, leaving little choice beyond environmentally questionable (e.g. diesel generators) or expensive-to-purchase (e.g. solar panels) power-generation technologies. It therefore makes sense on many grounds — operational, financial and environmental — to use the most power-efficient computers available.
Computer Aid International provides professionally refurbished computers for reuse in education, health and not-for-profit organisations in developing countries.
There are now a number of low-power computers on the market, ranging from notebooks (which are generally designed from the ground up with power consumption in mind), to small-format desktop PCs, to thin-client solutions that hang a number of minimalist workstations off a central host PC or server. Some of these products are explicitly aimed at developing countries, while others are more focused on the increasing need to conserve power in developed countries. But which type of low-power solution is best? Beyond the manufacturers' claims, there is very little independent information on this question.
In this initial survey, ZDNet UK has gathered eight examples of low-power computers, including high-profile products such as the OLPC XO and Intel's Classmate, and measured their power consumption when performing a representative set of tasks. For the battery-powered notebooks, we also measured rundown and recharge times. In addition to the lab tests, we have made a comprehensive log of each system's hardware and software features, giving extra weight to attributes that are particularly relevant to the developing world. Naturally, we also spent time simply using the computers, exploring their functionality and ergonomics. Finally, looking beyond the systems themselves, we have considered important factors such as the availablilty of documentation and technical support, and of training to help teachers, in particular, make the best use of the technology.
The outcome of this process is a shortlist of low-power solutions comprising at least one mini-desktop, one thin client system and one notebook. Six examples of each product will go on to the second stage of the project: field-testing in Africa, which will be carried out by universities in Kenya, Nigeria and Zimbabwe. The outcome of this trial, which will concentrate as much on 'ecosystem' factors as the technology itself, will be described in a later report.
What we tested
After surveying the low-power computing market and requesting some 15 products, we ended up with the following eight systems for testing in ZDNet UK's labs:
Lenovo ThinkCentre A61e* (compact desktop PC)
Read review
Aleutia E1* (mini-desktop PC)
Read review
Inveneo Computing Station (mini-desktop PC)
Read review
NComputing L230* (thin client)
Read review
NComputing X300* (thin client)
Read review
ASUS Eee (notebook)
Read review
Intel Classmate PC
(notebook)
Read review
OLPC XO (notebook)
Read review
For those products that did not ship with a monitor or have one built in (marked with * above), we used NEC's 15in. AccuSync LCD52VM, an XGA-resolution TFT display costing £105 (ex. VAT). For all power consumption tests (see next page), we set this monitor's brightness to 50 per cent, at which level it consumed 15 watts.
How we tested
Power consumption was measured using a Voltcraft Plus Digital Multimeter VC-940, obtained from Conrad.com. With the optional USB adapter and recording software, the data stream from the multimeter can be captured on a PC and saved in Excel format for further analysis.
The Voltcraft VC-940 digital multimeter, which we used to measure power consumption. Here, it's measuring the power drawn by an OLPC XO (with 100 percent screen brightness and with the battery charging) and delivering a data stream over USB to recording software on a Toshiba notebook.
Typical workload test
Our main benchmark was a typical workload test — 'typical' in this context meaning relatively low-end tasks using mainstream productivity applications that are likely to be undertaken routinely by shoolchildren in developing countries.
To this end, the test involved (1) typing a short (187-word) document into a word processor, (2) entering three columns of figures into a spreadsheet and creating a graph, (3) opening a browser and using Google to find the Computer Aid web site and then open it, and (4) going to YouTube and playing a video showing Nelson Mandela talking about the meaning of the word Ubuntu. From the resulting data stream, we extracted the following measurements: power draw when idle (at the OS desktop); average power draw during the workload test; peak power; time taken to complete the workload test; and total power consumed during the workload test (in watt-hours).
Output from the Voltcraft VC-940 digital multimeter for the Inveneo Computing Station running our workload test, under Windows XP (blue line) and Linux (red line).
The workload test was performed by a human rather than run from an automated script, so the time taken to complete the test is not purely a measure of system performance. However, we used the same human operator (CM), and on the systems where we performed repeat tests, we got a variability of around 7 percent on average.
Other measurements we recorded separately were standby power (with the computer plugged in but turned off), boot-up time (to the OS desktop) and browser launch time (to Google, set as the home page). Finally, for the three notebooks in the test, we recorded battery rundown and recharge times.
Power consumption figures measured in our workload test broadly follow the expected pattern, with the OLPC XO being the most power-frugal system (7.4W on average in reflective/greyscale mode) and the Lenovo ThinkCentre A61e, a relatively conventional desktop system, drawing the most power (46.4W on average, peaking at 73.4W). The other two notebooks, the ASUS Eee and the Intel Classmate, take well under 20W on average, while the mini-desktop Inveneo Computing Station hovers around the 20W mark (19.4W running Windows XP, 20.7W running Linux). The other mini-desktop we tested, the Aleutia E1, draws just under 3W more than the Inveneo system on average.
The numbers for the two NComputing shared-access terminal solutions need a few caveats attached. The biggest determinants of overall power consumption (and performance) are the specification of the host PC and the monitors used, as the access terminals themselves draw next to no power (1W for the X300, 5W for the L230). We used a pretty standard 2.8GHz single-core Pentium 4 PC from HP with 1GB of RAM and a 15in. XGA-resolution LCD monitor from NEC with brightness set to 50 per cent.
The power consumption figures reported here (average power of 33.2W for the L230, 29.2W for the X300) are per-seat for a system comprising the host PC and three access terminals. Although this is higher than the mini-desktops and notebooks we tested, it's worth bearing in mind that the host PC could support more terminals — ten in the case of the L230 (with Windows XP on the host, as here) and six for the X300. This would deliver lower power consumption per seat — but also, of course, put more stress on the host PC leading to slower performance as well as a higher probability of host PC failure.
Although we're mindful of the variability introduced by having our workload test performed by a human rather than an automated script, three performance-related points emerge from this chart. First, the 433MHz OLPC XO is well behind the two 900MHz notebooks (the ASUS Eee and the Intel Classmate). Much of the OLPC's extra time in the workload test was down to the YouTube video playback segment: the installed Gnash player wouldn't play YouTube video at all, but even when this was replaced by Adobe's flash player, playback was slow and jerky.
Second, the even lower-spec 200MHz Aleutia E1, which runs Puppy Linux from a 1GB CompactFlash card, is seriously underpowered. It took over an hour to complete a test that took 18 minutes at best (on the Intel Classmate). Applications are particuarly slow to load on the E1, and video playback is again slow and jerky.
Finally, when it comes to the Inveno Computing Station, it was noticeably quicker to complete the test under Windows XP than Linux (24.6 minutes compared to 31.1 minutes). This was largely due to faster application load times under Windows XP.
This chart shows the total power consumed during our workload test, and therefore provides an indication of the systems that deliver the best combination of performance and power consumption — at least with the mix of tasks we selected. So although the Aleutia E1 is a relatively low-power system, its poor performance means that it takes so long to complete the test that it ends up consuming more power overall than the Lenovo ThinkCentre A61e, which is the most power-hungry of the eight systems on test.
At the other end of the scale, the ASUS Eee and Intel Classmate — both very similarly specified 900MHz notebooks — emerge as the most power-efficient systems, consuming around a quarter of the power racked up by the Aleutia E1. The OLPC XO, although the most power-frugal design on test, was penalised by its slower speed, especially when playing back video, and emerged on a par with the mini-desktop Inveneo Computing Station rather than its notebook brethren.
As mentioned earlier, the NComputing products are very dependent on the host PC/monitor/terminal loading combination, and it's probably fair to say that the figures reported here could be improved upon with a different configuration. Also, a simpler workload test with less emphasis on video playback would show the Aleutia E1 and OLPC XO in a better light.
Conclusion
The purpose of this exercise was to investigate the power consumption characteristics of a representative collection of nominally 'low-power' computers. These must be suitable for deployment in schools and other organisations operating in rural areas in developing countries. We ended up looking at eight systems: one desktop, two mini-desktops, two thin clients and three notebooks.
Having carried out the workload test described in the preceding pages, and also reviewed each system in the traditional manner, we now need to select the products that will go on for additional field-testing by Computer Aid-affiliated universities in Kenya, Nigeria and Zimbabwe. We have decided, in the interest of further comparative enlightenment, to choose at least one mini-desktop, one thin client and one notebook.
Mini-desktop
Mini-desktop choice: Inveneo Computing Station.
The easiest choice to make is the mini-desktop, which is the Inveneo Computing Station. This is a genuinely low-power system (~20W average power) that performs reasonably well under both Linux and Windows XP, and is specifically designed for use in developing countries — in particular, it can run off a 12V DC solar power supply if necessary. Inveneo also has a network of partners in various African countries who are certified to provide support.
Thin client
Thin client choice: NComputing X300.
Although the NComputing thin clients delivered moderate overall power efficiency in our test, we feel that the PCI-based X300 product would repay further testing in the field. Many organisations in developing countries will already have suitable host PCs and should be able to source monitors, keyboards and mice reasonably cheaply. This means that the X300 will be extremely cost-efficient, given that the product itself costs just £149 for three access terminals and a PCI card. The X300 terminals, which can run from Linux- or Windows-based host PCs, are also durable as they are sealed units with no moving parts, and should be able to cope with testing environmental conditions.
Notebooks
Notebook choices: ASUS Eee, OLPC XO, Intel Classmate.
The ASUS Eee is the only one of the three notebooks we tested that's available on the open market. It also delivered an impressive combination of low power consumption and performance, and has not — to our knowledge — been formally field-tested in rural developing-world conditions. All of this makes the Eee an automatic choice in the notebook category.
The OLPC XO and the Intel Classmate are both high-profile products that are already deployed in various developing countries, and we initally considered leaving it at that. However, since these two notebooks are specifically designed for use in challenging environments, we felt it would be valuable to include them in the African phase of the project as well.
We look forward, then, to bringing you further details on the Inveneo Computing Station, NComputing X300, ASUS Eee, OLPC XO and Intel Classmate in the coming months.
Acknowledgements
Power testing Terry Relph-Knight
Computer Aid liaison Ugo Vallauri