Networking virtual servers: A long way to go

The tech industry is gaga over server virtualization and cloud computing. It seems like every vendor I speak with describes massive data centers connected to third-party clouds anchored by huge numbers of physical and virtual devices.

Sounds pretty cool, but I wonder how all of these physical and virtual distributed systems will be networked together? This is where reality becomes an ugly guest at the industry virtualization/cloud party.

When you look at how users are connecting physical and virtual servers and networks today, here is what you find:

1. Network design remains extremely simple. Typically, physical servers run 5 to 10 virtual servers. These virtual servers talk to each other through a virtual switch and talk to the rest of the world through a simple connection to a basic Ethernet switch. In truth, VMware ESX offers far more sophisticated networking capabilities, but since most virtualization projects are really focused on consolidating Windows workloads on a reduced number of physical servers, simple connectivity is good enough.

2. Skills are limited. So if you wanted to design a sophisticated network composed of virtual and physical devices with added functionality for security, performance, and high availability, who the heck do you call? Beats me. It seems to me that these skills will be in high demand and some organization will create a certification program down the line. But these future developments won't help today.

3. Tools are immature. OK, so what happens when the performance of a particular virtual server becomes a problem. How do you troubleshoot this? When virtual systems share physical resources, you better have tools that understand this relationship. Lots of companies claim that they have this problem licked, but users still tell me that they continue "flying blind" in these situations.

I haven't even mentioned security, which is another complex problem that the industry is glossing over.

Ultimately, we as an industry have little idea what will happen when you start networking massive numbers of physical and virtual devices together in a single data center, let alone a public network for connection to some type of cloud. We thought we could consolidate applications and serve them up over the WAN. Turns out we needed to develop dedicated technology called WAN optimization to get this right. I guarantee we will find numerous similar pitfalls.

Forgive me for being a cynic, but every time I get my head in the clouds, IT professionals, ESG Research data, and my virtualization-savvy colleague Mark Bowker throw a virtual bucket of cold water in my face. I am simply passing along the favor.

[MrJosh]

Well said, its a fascinating idea, but as with all things bleeding edge, we'll see a good deal of nicked fingers before the technology matures to the level marketing would like the masses to believe it has. I for one have been working on the VM side of networking for only a few short months really. The possibilities are as endless as the problems that arise from it. Here's hoping it comes along fast though, because as an industry we are going to be dealing with it on the behalf of clients now, whether or not its ready at the moment. Such is the fate of any industry driven by a market.

[jabberwolf]

This is exactly what we are running into.
And its not just 1 flavor but imagine having some vmware, some hyper-V (microsoft) and some on a SAN.
HP insight has done a great job but thats only if you are sticking with their hardware.

Its not really technology or 1 or 2 points of management, that's been licked. Its really instituting policy and procedures to ensure you can manage them correctly. And yes, there should be a better product for testing network I/O for servers, as sometimes I agree that I'm flying blindly and trusting that these new apps are telling me the truth.

[rapier1]

You mean putting 10 critical servers behind a single ethernet port doesn't provide you with a whole lot of redundancy? Surely you jest!

[lennyrvi]

Tools for virtual infrastructure performance optimization are now viable. One of the biggest problems users run into when deploying larger-scale virtualization projects is I/O performance. Understanding how a transaction being processed from one VM on an ESX server relates to a virtualized SAN infrastructure is very complex and extremely hard to debug when performance slows to a crawl. Without the right tools, understanding what is happening on the SAN is impossible. Check out www.virtualinstruments.com to see one approach to this problem that is being deployed by dozens of F500 accounts to solve this problem. Virtual Instruments does not solve all of the virtualization management problems, but for anyone concerned about performance of virtualized applications, it's worth a look.