The way things appear, from a networking stand-point the 3 key technologies that make or break cloud computing are data centers, Ethernet (in various manifestations) and optical transport.
Data centers are the brain that one can say is being outsourced! In order to facilitate availability of processing power across the web, we need network attached intelligence with added computational, storage and collaborative abilities - all of which are well juxtaposed by data centers. DCs can be of 2 types, the stand alone one or the distributed one, and given a good communicative infrastructure, either of the two should not matter. The intra DC design itself is a subject of tremendous introspection and research and a lot of work needs to be done here as well. Especially if one is to build these using server or even desktop farms. Which means that there are issues at higher layers like file sharing and file systems itself that need to be optimized. Yahoo's hadoop is a good mechanism to do the former and is well recommended by a large number of non-profit organizations.
The second key technology is in the transport of data between the DCs, the DC and the end user and between multiple businesses (DCs again). While not much attention is being given to this - relegating this area to the likes of equipment vendors, it should be sufficiently be clear that this area is an area of strong growth as the networking requirements are far different from anything we have seen before. Voluminous, latency sensitive, dynamic bandwidth needs cannot be met through dedicated circuits or unpredictable packets. We need something better, something with a packet nature but a circuit type reliability. Ethernet in the metro and long haul is likely to be such a candidate technology. Making Ethernet dynamic, reliable and available will be the key for facilitating efficient cloud computing across large networks. It is quite unlikely that the present variants of carrier Ethernet alone could scale, in fact, they might be best suited to be frameworks to be able to facilitate something more powerful - a new Ethernet pedigree. In our CAMPUS solution that we are presently building one such approach is being considered. But this may not be the only way. More on the CAMPUS solution in a few weeks from now.
Finally, the underlying transport technology is important for the cloud. From a stack perspective the physical layer is simply something that would carry bits. At least that is what most CS academics would tell ya. But the real thing is different - the phy layer can be an important weapon in the DCN arsenal. It is the single reflecting point in the network - allowing functionality to be cross communicated. The phy layer having the primary function of transporting huge chunks of data from one point to another, with minimal errors is naturally composed of optical fibers. But if we think about the added requirement on the phy layer, we realize that these optical circuits need to be reconfigured dynamically in two ways: (1) the optical circuits themseleves are to be set up and torn down as part of a fast moving virtual topology. (2) the circuits may have different line rates which change rapidly. Moreover these circuits are subject to particular data center requirements. The optical layer hence can play a key role in meeting these. Some of the technologies we have developed are well suited to this - see my CAVALIER paper.
I read this interesting blog on Ciscos website on DCNs and the way things are moving.
I was also following up on Nuova's acquisition (by Cisco) and what that means for the data center business.
People are getting this right but there is still a bit more to be made, and being bottom heavy may be the way to go.
