Virtualisation technology is expected to start moving onto production networks this year and next. How should network managers prepare?
We’ve dealt with physical-server sprawl and we know what that is. Now we have virtual-server sprawl. You can very quickly deploy virtual machines into an environment and it’s very easy to get virtual-machine sprawl. You can end up with capacity issues and resource allocation issues, so a lot of the initial challenges are around how to optimise the physical infrastructure for all the virtual machines.
What would help prevent this virtual sprawl from affecting network performance?
Getting visibility into the behaviour of the virtual machines that are running in production. If you think of VMware’s VMotion technology, which allows you to move live servers around in real time, there are still some manual processes for being able to identify virtual machines that may be performing badly and trying to correlate that with how the physical servers are performing. IT managers need to orchestrate the alerts and then be able to move virtual machines to the right place at the right time to optimise performance and capacity.
Why won’t the tools or processes network managers use today stand up in this virtual environment?
It would just be impossible based on the number of alerts — which could be hundreds to thousands of alerts on performance issues. To be able to manually reallocate resources is just not feasible and it goes against the whole value of virtualisation. You would end up having to provision a bunch of overhead so that you could move the virtual machines around successfully without interrupting anything. And that defeats the whole purpose of trying to optimise the infrastructure.
Will virtualisation change the way vendors have to manage infrastructure?
Absolutely. A lot of monitoring today is still infrastructure-focused. It focuses on static thresholds that are set on the physical devices. And there are still a lot of manual processes that take place to understand and correlate the performance of virtual machines with the performance of the physical machines.
Is automation the best option to handle virtualisation in today’s datacentres?
Datacentre administrators are still very nervous about turning the keys over to an automation engine. But if they can get a baseline of behaviour, and get the same alerts over and over again saying, “Are you sure you want to do this?” they can eventually click the little box that says don’t ask me any more and trust the product to do it.
What if any impact does such server virtualisation technology have on the network?
That’s the thing. If you get into a full production environment, and you have virtual machines that are being reallocated and reprovisioned on the fly continuously, there will be an impact to the network. The switch architecture is going to have to have some visibility into the application traffic. Moving servers around within the datacentre will increase variability on the network itself.
How will vendors have to innovate to tackle virtualisation?
What’s really next and the emerging growth area in management is around service orchestration. A user makes a request, application components get bound together based on that request, and infrastructure needs to be associated to fulfill that request. That is where a lot of research and development is being done, especially by IBM, HP, CA and BMC. They are looking at how to dynamically allocate resources per policy or when behaviour is a certain way. A lot of the innovation is going to be in that level of orchestration.