Thursday, November 13, 2008

Arista Networks Take 2



Turns out that Andy Bechtolsheim was giving a talk on Arista here today. Luckily I got a last-minute notice and squeezed into the room. The talk was divided into two parts - cloud computing and cloud networking.

In the first part, Andy reviewed the remarkable market size and growth of cloud computing, and how customers and vendors view it. It is estimated that the cloud computing infrastructure size will reach a staggering $42-billion by 2012. There is certainly much incentive for them to be a big part of that market!

Customers would love cloud computing because services will be (expectedly) cheaper, faster to obtain, and simpler to use. There is no upfront investment to get an ugly box with CDs/DVDs, and no need to worry about losing the discs, since the service is accessible anywhere at anytime. From a vendor's perspective, cloud computing provides a cheap way of making services available to a large audience. There is no infrastructure cost apart from renting the hardware resources in the cloud. It is also possible to continuously add stuff to services without having to ship out patches / upgrades.

There are, however, certain reservations about cloud computing. For instance, customers might not want their data to be put on the potentially insecure cloud, and could have trouble adopting new applications in their daily routine. Vendors might also have inertia in moving services to the cloud since that may cost them their large user base. There is also the issue of agreements on responsibilities should hardware or software fail in the cloud. In essence, it may not be clear who should be responsible for what kinds of failure. Revenue is another concern, where vendors may have little profit in providing per-use services.

In the second part of the talk, Andy explained the necessity of cloud networking equipment. Basically, the cloud infrastructure is different from that of existing enterprise networks. There will be tens of thousands of servers providing a non-blocking, low-latency multi-terabit capacity and supporting dynamic on-demand application deployment. It must be reliable all-year-round and be of low cost.

Arista's short-term target is to achieve a $100 server with 1Gbps non-blocking capacity. But the real challenge is 10Gbps and up (40Gbps). They envision that 10 gigabit Ethernet would become mainstream in 2009, and almost all motherboards would have a built-in 10 gigabit Ethernet adapter by 2012. The increased capacity brings about more possibilities in designing cloud applications. A nice direction for tech companies is, then, to think of what kinds of applications would be feasible and popular in the next 5 years.

0 comments:

Archive