Citrix, Juniper, HP, Yahoo and Nicira on the future on networking in virtual infrastructures

Virtualization and cloud computing are changing the way we design data centers. The more powerful CPUs Intel and AMD produce, the more virtual machines per core administrators can host on a single hypervisor. But the higher consolidation ratio we achieve the more issues we have with memory, storage and networking components, that are quickly becoming the new virtual infrastructure bottlenecks.

Virtualization vendors try to overcome memory limitations with several overcommitment techniques, like the new Memory Compression from VMware and the upcoming Dynamic Memory from Microsoft, while storage vendors try to develop more virtualization-friendly SANs controllers able to facilitate acrobatics like long-distance virtual machines live migrations, like the EMC VPLEX
Excluding Cisco and HP, established networking vendors don’t seem equally busy in addressing the new challenges that exist in virtual and cloud computing infrastructures.

This topic has been covered a number of times before. The last one is in a roundtable hosted by the Association for Computing Machinery (ACM) last month.

The group of experts that participated the round table is particularly interesting as it includes CTOs and Vice Presidents from Citrix, Juniper, HP, Yahoo! and even the semi-stealth startup Nicira, where the founder and former CEO of VMware Diane Greene invested.

There are a lot of interesting comments that help to understand where these companies are going or at least how they view the challenge.

From Citrix (with our emphasis):

…given the progress of Moore’s law and the large number of VMs (virtual machines) we can run per server, the implicit change to networking is that the last-hop switch is necessarily a feature of the hypervisor or hardware of the server and not a traditional hardware switch in the physical network.

IaaS challenges the traditional vendor/customer roles for networking equipment. It may be that the cloud vendor purchased equipment from a specific vendor, but there is no way for that vendor to surface its unique value proposition to the IaaS customer. Does this necessarily force commoditization in network equipment? I think it does. Google, for example, reportedly already builds its own networking gear from industry-standard parts.

The key point is that you don’t have the luxury of being asked when a VM moves; you are told. The argument that [HP] makes is that we would never move a thing to a LAN segment that is not protected. People usually don’t understand the infrastructure at that level of detail. When the IT guy sees a load not being adequately serviced and sees spare capacity, the service gets moved so the load is adequately resourced. End of story: it will move to the edge. You are not asked if the move is OK, you are told about it after it happens. The challenge is to incorporate the constraints that Lin mentions in the automation logic that relates to how/when/where workloads may execute. This in turn requires substantial management change in IT processes.

From Nicira:

Originally, the leverage point was in the network because it was central. Because of this, networks have always been an obvious place to put things such as configuration state. Now the leverage point is at the edge because the semantics there are very rich. I know where a VM is, I know who’s on it, and I know when it joins and when it leaves. As a result, I don’t require traditional service discovery and often don’t need multicast. Because the leverage point is at the edge, the dynamic changes completely; and because the semantics are now more interesting at the edge, you have a clash of paradigms.

From Yahoo! (with our emphasis):

In the next two to three years our goal is to make the building of an application, its packaging, and deployment completely transparent. I want to specify SLA (service-level agreement), latency, and x-megabit-per-second throughput and receive a virtual network that satisfies the requirement.

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.

From Juniper:

…At Juniper, we want to build what is in effect a stateless, high-capacity, 100,000-port switch but without backhauling everything to the “god box” in the middle.

Surprisingly, a networking vendor that has much potential in the virtualization marketing, Vyatta, was not part of the discussion.  
Despite that (and the lack of Cisco too), the roundtable is extremely interesting and definitively worth a read.