A few years ago I wrote an article on computer cooling technologies. That article focused on various ways to reduce the heat emitted by computer equipment. It also included a section on data warehouses, or server farms.
Regardless of the cooling technologies used there remains a significant amount of heat that needs to be dealt with. Even a modest data center can easily emit temperatures around 40º-50º Celsius. Most often, data centers employ HVAC systems which attempt to mitigate the heat. The downside to this is that it’s a loosing battle. You’re using a lot of electricity to power the data center which gives off heat. You’re then using even more electricity to combat the heat.
Recently Microsoft researches have proposed to re-use this heat. Instead of fighting the heat they want to create data furnaces which will use the heat to heat homes and businesses. Their idea is to sell fully enclosed data furnaces much like a regular furnace. The data furnace would physically fit into a basement or closet and integrate into an existing duct system to distribute heat. Additionally, the data furnace would be connected to the Internet and be secured from tampering.
The Internet Connection
One of the major factors in the idea of a data center is the network connection. Data connections in today’s data centers are typically at 10Gbps, although this is shared amongst hundreds of thousands of servers. If a data center were to be replaced with thousands of data furnaces distributed to businesses and homes could the networks provide adequate service?
One thing to note is that not all cloud-based computing services require high performance networking. For example there are plenty of non-time sensitive computations required by the scientific community, SETI@home for example. Microsoft researchers also give the example of non-real-time web crawling and content indexing tasks. These computations could easily be started and stopped as the data furnace is required to produce heat (at night for example) and transferred to a different data furnace as required.
Many businesses or even some homes may already have, or be willing to upgrade to, a faster Internet connection. There are services available to residential customers today that can provide 70Mbps download and even 30Mbps upload speeds. While these speeds still don’t compare to the 10Gbps speeds of data centers they would not have to be shared with as many servers (perhaps a few hundred along with the data furnace hosts). Also, the data furnaces could be distributed around a highly dense population and could provide content quickly to this population as the geographical distance is negligible.
If a data furnace can reliably produce enough heat to satisfy an average family home in a cool climate - what are the ramifications? For one, the household would no longer need to heat their home with gas or oil. This could save some money but would be offset by the added electrical costs for running the data center.
Unless you live in the arctic you probably don’t need heat year round. In the summer, you’d likely not need any heat. While in the fall and spring you might only need heat during the night time. During the winter months is likely the only time some would need heat all day and night long. Microsoft suggests that the data furnaces could still run while the host does not require heat, and that the heat could be pumped outside as long as the temperature is less than 35º Celsius. When the outside temperature is greater than 35º Celsius the data furnace would need to be shut off.
So if we accept that each data furnace host would have a different profile of when the data furnace could operate and at what capacity then there are two solutions we could employ. The first solution would be to find applications that are tolerant of these heat profiles. As mentioned earlier complex scientific calculations would be an appropriate use for data furnaces that could only run at night or at reduced capacity at any given day depending on the temperature.
A second solution would be to have many hundreds of thousands of data furnaces geographically spread around the globe. This way, when it’s summer time in the northern hemisphere (and many data furnaces would be shut-off during the day) it would be winter time in the southern hemisphere (and data furnaces could work at full capacity). Also, if a data furnace needs to be shut-off during the daytime in Boston it would be nighttime in Berlin where other data furnaces could be running. So if we have enough data furnaces with adequate geographical dispersion we could virtually always have a large percentage of the total number of data furnaces working. The challenge in this solution would be to adequately transfer the workload around the globe to the appropriate data furnaces. This can easily be accomplished as long as the computations support this type of transfer; the previous example of scientific calculations would likely fit this use case as well.
The researches at Microsoft point out some pretty compelling financial benefits of data furnaces. This is largely due to the fact that data furnaces would be housed in an existing home or business. So there is no new building costs as there would be with a traditional data center. Even if you factor in more expensive electrical costs (due to the data furnaces being in a residential area as opposed to a less expensive industrial area) the total operating costs (per CPU) of a data furnace is considerably less than the operating costs of a data center.
Another advantage of data furnaces is that since heat is a valuable by-product then the furnaces can be made up of older, less-efficient servers. In a traditional data center, it often makes sense to replace servers with newer, more efficient models, to save on the electrical and cooling costs even though the hardware itself is still capable. These replaced servers could be purchased at a discount to be placed in data furnaces.
The researches at Microsoft do an excellent job promoting the case for data furnaces, from a technical perspective. I can certainly see that data furnaces could be computationally adequate for many uses. I can also imagine that they could be adequate heating sources. But would home and business owners care to switch their existing heating source for a data furnace? What’s in it for them? The operators of data furnaces would certainly save money compared to operating a data center and let’s assume that they could charge a comparative amount for their computational use. This is what I see as the biggest obstacle in this idea becoming a reality. There would need to be a compelling business plan that would encourage people to sign up for this. Basically, I think people would need to save a significant amount of money in order to host data furnaces.
Alternative, if the data furnace host could somehow take advantage of the data furnace then perhaps some businesses would use them. For instance, if the data furnace (in addition to performing it’s regular duties) could also positively contribute to the host business’s computing needs then perhaps this could be of significant value to the host.