Monday, September 7, 2009

HP ProCurve's BattlePlan

Paul Congdon is chief technology officer (CTO) at ProCurve, the division of HP that manufactures switches, routers and other networking products. He talked to Computing about how network virtualisation can improve the performance and manageability of virtual servers in the datacentre.

Analyst research suggests Ethernet switch sales are down 24 per cent year on year - how can ProCurve get companies buying again?

The market has been down overall, but we have had enormous success with wireless, seeing 112 per cent growth even though the market was declining by 20 per cent overall. We are closing the gap on Cisco in sales of 10Gbit/s Ethernet equipment as well.




Many organisations still find it difficult to obtain cash or get a loan, while ProCurve has been traditionally strong in the public and education sector, and in the US at least government cutbacks have presented a challenge – public sector money is not as available as it used to be. We are offering various incentives within the [reseller] channel, trade-in programmes and initiatives around lead generation. There are no heavy discounts, but the margins have not changed much.

What new technology will tempt firms into upgrading their existing network infrastructure?


Embedding security in switches is an ongoing process. One of those things with the whole Moore’s Law process is the density of silicon and the amount of space within chipsets to add programmable functions to switches. The other is network virtualisation, where we can create virtual ports inside a switch. You might have 24 physical Ethernet ports, but another 200 virtual ports, for example.

What will network virtualisation do?


There are a couple of approaches to the technology being proposed to the Institute of Electrical and Electronics Engineers (IEEE), one backed by HP ProCurve, and another by Cisco. The driving force here is that a virtual server has a piece of network infrastructure, a virtual switch, embedded in its software, which could actually be moved into hardware to improve virtual server performance and manageability.

The big questions are what should the virtual switch do, how much number crunching should it implement, and where should it reside? Network managers generally like to configure devices to make them behave the way they want them to, either within virtual software on the server, or at the edge of the network. Either the servers need to take on new functions, such as firewalls, or that stuff has to be put out on the network where the switches can do it for them.

What would you recommend as the best approach?


HP ProCurve proposes choice and flexibility, a way to configure virtual software so it can direct traffic out onto the edge of the network rather than have every feature known to mankind embedded within software that will ultimately run on a network interface card (NIC) with input/output (I/O) virtualisation capabilities.

The difference here could be NICs that cost $500 rather than $5, so we are very interested in finding a way that uses the capabilities in existing switches so the upgrade path is less disruptive. We could add a lot of that software into a switch without the capital cost of the NIC, it just comes down to a question of complexity and management – configuring firewall rules between new virtual LANs (VLANs) for example, could be done on one switch or every switch in the system, and that could create management problems for some people. There is trade off between performance and functionality and customers should be able to choose between the two rather than be limited to what is put inside NICs.

What are the potential benefits of network virtualisation for the IT manager?

Performance and manageability are foremost, but it is also ease of migration. You could import a base set of these capabilities into a switch, for example, without requiring a hardware upgrade, and therefore experiment without taking on capital costs right away. There are some open-source solutions available now – I wrote some of the code myself.

Where will demand for this sort of network virtualisation come from?


Datacentres will be the big beneficiaries, but also imagine how that extends to wireless environments. If you can embed a wireless controller in an Ethernet switch, you can use virtual ports to represent wireless users, then terminate virtual private network (VPN) connections on virtual ports [without taking up physical ports on the switch].

No comments: