* Storage over IP requires no expensive switching components
Today, another installment in my series on the Human Speechome Project that’s now getting underway at MIT’s Media Lab.
As you may recall from my earlier articles, the project is interesting to storage weenies for several important reasons. First, the project forms a storage grid that, while starting small, is adding data at a rate of about a terabyte each week, eventually topping off at 1.3 petabytes by the end of its three-year run. Any storage environment measured in terms of petabytes is interesting, even if only because the technology that runs something that large today will be running much smaller environments tomorrow.
Second, the grid being formed is built from relatively inexpensive off-the-shelf products you can buy today (I began an assessment of one of these boxes in one of last week’s columns). The disks are SATA; the connectivity is Gigabit Ethernet.
Third, the project creates a storage-area network built on a new technology, Storage over Internet Protocol (SoIP), which requires no expensive switching components – in fact, it uses no switches at all – and relies on the same readily available IP routing technology that you use at home. The folks at MIT will be using something much more robust I am sure, but I have one of these boxes hooked up to a $49 NetGear router while I do my testing.
Finally, the whole thing is being architected to be run by an IT staff with no background in data storage. Admittedly, this “staff” is not really typical of what you are likely to find in most computer rooms – these are graduate students (and MIT grad students at that), but they bring no particular storage expertise to the table, and their involvement in this project is based on interests that are not storage-related. For them, storage is what it should be, a means to an end.
The grid will be comprised of a number of storage arrays set up for RAID 10 (or RAID 1+0 if you prefer – I do, I think it’s more descriptive). One interesting aspect of the SoIP technology is that its IP connectivity means that RAID partitions define disk relationships based on IP addresses and not by cabinetry. It is thus theoretically possible for a volume to be physically global in nature, although the limitations imposed by long-haul traffic over IP probably mean that however possible that may be, it is unlikely to be practical (darn that speed of light thing).
I’m keeping an eye on this for the foreseeable future, and will continue to report on anything interesting I uncover. In the meanwhile, if you want to see more on this, see the video of MIT Media Lab’s press event announcing the project.
