cycles enough to make it worthwhile to burn switching cycles inside the server.

If I’m a network guy in IT, I better learn the concept of port groups, how VMware, Xen, and others work much more intensely, and then figure out how to get control of the password and get on the edge. Those folks now have options that they have never had before.

The guys managing the servers are not qualified to lead on this because they don’t understand the concept of a single shared network. They think in terms of bandwidth and VPLS (virtual private LAN service) instead of thinking about the network as one system that everybody shares and is way oversubscribed.

ReDDy: We are moving to Xen and building a new data center architecture with flat networks. We tried to use VLANs, but we have taken a different approach and are going to a flat layer 2 network. On top of this we are building an open vSwitch model placing everything in the fabric on the server.

My problem is in responding to the service requirements of my applications and addressing things such as latency and throughput. The data needed to address these issues is not available from either a network virtualization solution or the hypervisor.

Also, my uplink from the switches is 10 gigabits per second or multiple 10 gigabits per second, but my NICs are only one gig. If I run 10 VMs on a box, then all of the bandwidth aggregates on one or two NICs.

neaSe: You guys are cheap. If you went to a backplane, then you would get a lot more bandwidth out of those servers. A KR signal on a backplane is how you get a cheap copper trace for 10-gigabit service.

caSaDo: Going forward, a new network layer is opening up so you can take advantage of virtualization. Traditional networking vendors certainly do not control it today and may not control it in the future. The implications are that it may not matter what networking hardware you purchase, but it may matter much more what network virtualization software you choose.

If you like the cost point and the service and operations model of the cloud, then look at Eucalyptus, Amazon, Rack-space, and so forth, and see how they build out their infrastructure. Today

mache cReeGeR
mass interest in
virtualizing the data
center is breaking
a lot of traditional
physical versus
logical bounds. you
need to concentrate
on what you’re
trying to achieve
in your data center
and what key
properties you’re
trying to preserve.

that is the only way you can get these types of flexibility and per-port costs.

It would be interesting to compare a vertically integrated enterprise with something like Amazon EC2 (Elastic Compute Cloud) in terms of cost per port and efficiency.

BeeLeR: The guys who run infrastructure for Google told us that the difference between running their own infrastructure and running their stuff on Amazon was small enough that it really made them think about whether they wanted to continue to do it themselves.

caSaDo: We have seen close to two orders of magnitude difference between a vertically integrated solution and something like EC2.

BeeLeR: The relevance here is that while these issues may not affect you today as a practitioner, you should understand them because they will affect you tomorrow. In this way you can make intelligent investments that will not preclude you from taking advantage of these kinds of benefits in the future.

caSaDo: The leverage point for vertical integration has always come from the networking vendors. It was lost a long time ago on the servers. Someone who offers a full solution is going to be a networking vendor. If you’re making a purchasing decision, then you don’t have to blindly follow the legacy architectures.

I do not believe that owners of existing network infrastructure need to worry about the hardware they already have in place. Chances are your existing network infrastructure provides adequate bandwidth. Longer term, networking functions are being pulled into software, and you can probably keep your infrastructure. The reason you buy hardware the next time will be because you need more bandwidth or less latency. It will not be because you need some virtualization function.

TaVakoLi: We get caught up on whether one is implementing a new data center from scratch or is incrementally adding to an existing one. For a new data center, there are several things to keep in mind. Number one, if you’re planning for the next five years, understand how you are going to avoid focusing on “rack versus row versus data center.” Architect the data center to minimize location-depen-dent constraints but still be able to