2011 Prediction: Netronome

It seems like yesterday I was compiling my thoughts for 2010 predictions, and now here we are looking forward to 2011 and the buzz for one technology seems to be prevailing over above all others… cloud computing.

A recent IDC report estimated the market for public cloud products and services will grow from $16 billion of IT spending in 2010 to $56 billion by 2014. Based on these numbers, one could say cloud computing is being under-hyped rather than overhyped.

Although there are many applications where the cloud is scaling well, many enterprise applications cannot be used on public clouds.  There are still too many kinks to sort out in terms of latency and security for the cloud to be ready for mass consumption. The benefits that a cloud offers are obvious: flexibility; elasticity; and, cost-savings. 2011 will see great changes and strides in the two keys areas: security and latency (there are many other areas such as management etc. that are important but these two are of particular interest).

Latency in today’s cloud systems, based on commodity Ethernet fabrics, can be compared to conducting a phone call over a satellite link where the latency affects the quality of the call. What we need is the equivalent of optical networks in cloud data centers where latency is much less.

Prediction #1 - Latency

Many of the cloud applications deployed today are first-order applications and are what I call “low-hanging fruit.” The next generation of applications need more intelligence (security), higher speed and lower latency communication capabilities.

For 2011 to be the breakout year for enterprise applications in the cloud, these applications need to be shown to scale in the cloud. A typical shared characteristic of many of these applications tends to be a shared database where synchronization of transactions is difficult, impacting the scalability of the cloud. For instance, if an IT manager is updating data in California and another member of IT is updating the same data in Florida the system needs to assure consistency which requires low latency. The higher the latency the fewer processors in the cloud can be used; adding processors actually slows down processing!

Minimizing latency in the cloud be done in a variety of ways - deploy expensive Infiniband switches, design custom switch fabrics that allow for low latency, or deploy more intelligence in the networking layer allowing intelligent optimization to occur.

Prediction #2 – The Cloud Requires a Specialized Processor

Latency is just one of the issues putting a hinder on cloud deployments. Another concern, which has been prevalent since cloud computing was introduced has been security. In 2011, we will begin to see more discussion around the importance of solving these top two concerns by addressing this at the network level.

A security mechanism deployed at the network edge is already moving into virtual deployments. The performance of many of these applications is however not sufficient to be used at the fine-grained level required and these applications need specialized, high- performance functions such as crypto and intrusion detection to be supported by dedicated silicon inside the server. A way to think about it is that almost all PC’s contain a graphics card to do the graphics - you don’t use your x86 to do your graphics. In the same way there needs to be silicon deployed in servers to optimize security and network processing.

If for example we had the performance to encrypt every data transfer - to or from the user, the disk, the network with a customer specific key - this could remove a very significant barrier to using the cloud.