The UMD: How Battery Performance Can Differ, Observations and Recommendations

by Lynn Merrill

With the introduction of Ultra Mobile Devices, the possibility of taking your work and entertainment everywhere in a very small form factor is a very appealing option for many mobile users. The possibility of having a complete and functional operating system and the accompanying software capabilities will make this new offering a tempting solution for those that are on-the-go. Naturally one would wonder, “Yes, I now have a very small platform that will do what my larger laptop or desktop machine will do, but what can I really accomplish on this device, and for how long?” This paper will discuss this important issue, look at some performance data with regard to battery life, and make some recommendations for preserving battery life to those who may be delivering new or modifying existing software for this platform.

The scope of this paper will be to discuss differences and commonalities between laptop and UMD platforms. First we will review techniques for extending battery life that are currently in use in the industry. Next we will examine and discuss importance differences between laptop and UMD platform components that can contribute to differences in battery life. We will then look at battery performance observed while performing a common workload on an IBM T42 laptop and a Samsung UMD platform. Finally, we will list some general observations and recommendations for development of software for the UMD platform.

For quite some time now, work has moved forward in making laptop computers run longer on batteries so that more work or enjoyment can be gleaned from the computer. Various methods have been tried, tested, and proven to be very effective in prolonging the life of the battery. Such methods include turning off features in applications when running on battery, caching data into memory to cut down on the drain that DVD readers and hard drives put on the battery, balancing block sizes on file operations, varying the interrupt rate on network and other operations. These and many more methods have allowed applications to run smarter with respect to battery usage and have allowed the user to accomplish more while away from the office in a disconnected setting. With the advent of the UMD and its ability to run a full operating system and accompanying software, it is imperative that application developers continue to optimize for power when creating their software products. As new and continuing research evolves, new and important techniques that will extent battery life are sure to be discovered and shared with the development community. These new methods should then be used where appropriate in extending the power usage on all mobile devices.

In just looking at a UMD device it is obvious that there are many differences between the form factor and the traditional laptop computer. Some of these differences will affect how the battery is utilized while in operation, while others will present new problems with which to contend. Let’s examine a few of these differences and their impact:

Screen Size
This key difference may provide the largest difference in battery usage on the UMD. For some time it has been known that the screen is the one of largest consumers of battery power on the mobile device. The smaller size will produce less drain on the battery and will help to prolong its usability. While the smaller screen may be a big plus for battery savings, it presents its own set of problems for application developers. With the greatly decreased display area, items such as UI elements, display windows, scroll bars, rendering windows, scroll bars, message boxes, font size, picture size, just to mention a few, will all have to be re-tested and perhaps re-engineered to ensure that the user experience is not compromised on the UMD. Great care will have to be taken to make sure the UI components adjust automatically to the capability of the display.

DVD Drive
As you look at the UMD you will note that there is no internal DVD Drive, although many UMD’s will come with an external USB based drive as an option. As another battery intensive device, the absence of this device may improve battery life vs. playback on a notebook. However, applications that heretofore used the DVD drive as the source of data will now rely on looking to the hard drive or over the network or via streaming technology to retrieve content. This may impact the way IO is read from or written to a device. Though this is great news for battery life, the absence of this device on the UMD many present different challenges.

Processor Speed
Current and upcoming UMD’s have a processor speed of around 800-900 MHz, whereas a typical laptop may have a processor whose speed reaches 2 GHz or higher. This factor will also cause a dual effect on the platform. A slower processor means less power is demanded from the battery as long as the processor utilization is low, thus extending the life of the battery. For most applications this will not affect performance either as most do not tax the processor to its maximum usage. However, for those applications that utilize the processor to 80% or more of its capacity, a drop in performance will be noticed. Perhaps games will not respond as quickly or refresh at an eye-popping rate, or other applications will seem to take longer to perform operations such as I/O, conversion, creation of graphical content, etc. Processor speed thus helps battery usage by using less, but may take some of the sparkle out of applications when compared to performance on a laptop.

Intel SpeedStep® Technology Impacts
Intel SpeedStep® technology (SST) was introduced to allow the processor to automatically vary its frequency and voltage based on the performance needs of the applications that are running at any given time. When the need for greater performance is high, SST boosts the frequency of the processor to a level, thus increasing performance. When the need for processor power is low, SST provides for a lower frequency. On a 2 GHz laptop the span in processor speeds can be 1.4 GHz, from 600 MHz on the low side and 2 GHz on the high. As you measure battery capacity and drain rate, the slope of the drain curve is much steeper at high frequency than when at a lower frequency, which will effect the battery life. But on the UMD the current span is only 300 MHz, from 600 MHz to 900 MHz. This serves to provide a much narrower range in speed differ ential and thus slows the drain rate of the battery. It does affect performance as well in that reduced performance may be observed. We will discuss this aspect in more detail in the observed performance section below.

Battery Capacity
Another aspect of battery performance is the amount of power that can be held by the battery when fully charged. On a laptop the typical battery will charge to hold some 32,000 milliwatt hours of capacity. For extended size batteries this number is much higher, which will allow for longer periods of usages. On the UMD device the battery will charge to hold only about 24,000 – 25,000 milliwatt hours of capacity. You can readily see that the UMD battery can only hold about 2/3 of the power of the typical laptop, thus giving it a sizeable handicap right from the start. Thus it should readily be observed that the other factors discussed here play a very large role in extending the life of the battery on the UMD.

For the purposes of covering this topic, we created a workload that was representative of someone who would use the computer either at home or at the office. Most people in both environments will use email to communicate with friends, family, and co-workers. Most will also attach pictures, documents, presentation foils, and the like to those messages. Thus our workload was created to imitate this activity. All files and documents that were used for our test were located locally on the hard drive of each machine, thus there is no network latency issues to skewed time while creating the email messages. Both machines were attached to the same isolated network that facilitated the creation and sending of the email messages. The same workload was run on an IBM T42 2.0 GHZ machine with 512 MB RAM and on a Samsung 900 MHz UMD machine with 512 MB RAM. Each machine was configured with Windows* XP SP2 with a power setting of Laptop/Portable. Battery data which was captured from the UMD device is plotted on Figure 1 below, while data captured from the IBM laptop is plotted on Figure 2. For brevity sake we will examine only the first 250 samplings of the data captured. While this represents only about 25% of the full workload, it is representative of what happened on both platforms during the entire workload. Common elements of both graphs are as follows:

Let us now do some analysis of the data presented. The first part of the workload causes the %TotalProcessorTime to spike to high usage rates. This part of the workload attaches items, such as pictures and documents, to email messages and then s ends the messages to like recipients identified in the mail message. As can be seen, both machines have very similar %TotalProcessorTime ranges during this workload. Several differences between to two machines can be observed:

• Two of the differences can be looked at together. The Discharge Rate in Figure 1 fluctuates much less that the rate in Figure 2. At the same time the frequency in Figure 1 remains constant, whereas the frequency in Figure 2 vacillates more widely. Raw data reveals that the frequency in Figure 1 remains constant at 897 MHz regardless of what usage the processor is using, while the frequency in Figure 2 ranges from 597 MHz to 1993 MHz depending on what processor utilization is being called for. This is an indication that Speed Step Technology is not as effective on the UMD device as it is on the laptop. The sensitivity of SST may be greater in the laptop than on the UMD, or it could be that SST is disabled on the UMD processor. The effect on battery performance is good news for the UMD, but it could be even better if SST worked to reduce the frequency on the UMD as it does on the laptop.
• The length of time to process the same workload in Figure 1 is longer that in Figure 2. If we look at the first sustained level of %TotalProcessorTime, Figure 1 runs approximately from time element 25 to time element 41, while Figure 2 runs approximately from time element 17 to 25. This helps valid the statement above that with lower frequency you will get degraded performance. It saves on battery power, but it does cost more in time. If you look at the full 250 time elements you will see that Figure 2 advances well beyond the point to which Figure 1 reaches at the end of the charted data.
• The change in battery capacity was an interesting observation. During the first 250 sampling capacity in Figure 1 decreased from 23,687mWh to 21,978 mWh. This calculates a slope of -6.62. During the same sampling time the capacity in Figure 2 decreased from 31,910 mWh to 28,920 nWh. This calculates a slope of -11.96. So the laptop consumes more power, but it performs the work at a faster rate.

Figure 1. Plot of Battery Data from UMPC

Figure 2. Plot of Battery Data from Laptop

The age old question of “Why should we take into account the battery usage when developing our application?” is often asked when examining new functionality or features, and determining whether or not it is relevant to a particular product. In the case of battery life, more and more people, both in the business environment and at home, may be purchasing UMD devices as their primary system. Applications will need to pay particular attention to battery performance to enhance the experience these new users will have on the UMD. Nothing can be as frustrating as being in the middle of a critical operation and have the computer shutdown for lack of power, especially if no warning is given alerting the user to a low battery level. As applications are either created or modified for corrections or for adding new features , software architects and strategists should include better and more robust power aware features by designing their software to help overcome the challenges of the mobile workforce and lifestyle.

We list here some typical use cases with recommendations that should help in providing for extended battery life, especially for the UMD:

One additional consideration may be to provide a custom defined power profile for your UMD users. The profile can be created to use the lowest power settings, the slowest processor speeds, and the most power miserly sleep states in an effort to extend usability of the computer. This relatively easy to create feature may save the user a lot of grief.

Exciting times are coming to the mobile world by way of the Ultra Mobile Device. With the power of a full size laptop or a desktop, these devices will bring to your fingertips in a very small form factor a full featured operation system and the accompanying suite of software packages you are already accustomed to using. That experience can only be fully enjoyed if there is ample power to facilitate an extended period of usage away from the wall socket. It is therefore imperative that as considerations for power optimizations are weighed and debated within the development community that great effort be taken to explore ideas that support extending the life of the battery. The success of this new platform could very well hinge on the manner in which outstanding performance is delivered with minimal power consumed. It is a great challeng e, and yet one that must not be ignored nor pushed onto the dusty shelf entitled ‘Someday’. The time is now to explore and embrace every means available to make the mobile experience the best that it can be.

For more complete information about compiler optimizations, see our Optimization Notice.

Comments (0)

Trackbacks (0)

Leave a comment