Arthur Cole spoke with Shehzad Merchant, senior director of strategy, Extreme Networks.
Cole: A lot of companies, yours included, are looking at 100 GbE as a fabric for cloud services. Please tell me that the cloud won't require that kind of jump -- even 40 G would be quite a stretch.
Merchant: In general, we view that any transition from a physical to a virtual to a cloud infrastructure will occur in a phased and planned migration. Currently, many data centers today are still operating at -- or upgrading to -- just 1 gigabit connectivity to the servers, with 10GbE running in the aggregation and/or core of the cloud fabric.
"... there are a couple of key trends emerging that may cause a potential shift to higher speeds for the fabric. One is the adoption of server virtualization, which leads to significantly higher traffic on each server."
However, there are a couple of key trends emerging that may cause a potential shift to higher speeds for the fabric. One is the adoption of server virtualization, which leads to significantly higher traffic on each server. The other is the commoditization of 10GbE connectivity. As these two trends converge, we think that servers will start moving towards 10GbE connectivity directly into the network. In turn, the data center fabric will start moving toward a 40GbE infrastructure. Concerning the timeframe, this would be a 12-to-18-month trend. Today, Gigabit Ethernet connectivity from servers into a 10GbE fabric is the predominant approach. It is important that IT infrastructure administrators making purchasing decisions today look very closely at the migration path and upgradability offered by their vendors, not only from a speed, density and scaling perspective, but also from the ability to deal with virtualization in a open and non-proprietary manner.
Cole: How should IT infrastructure itself change? Are we all destined for giant core switches that must be managed and maintained like a mainframe?
Merchant: Today the network infrastructure in the data center particularly seems to be moving toward a 4- or 5-tier model. You have your traditional core, aggregation and access layers, but in addition you now see a virtual switch tier as well as a blade switch tier. That's clearly too many tiers. Each tier introduces latency, oversubscription of bandwidth and management overhead. So there needs to be some collapsing and consolidation of the tiers to reduce the latency and oversubscription, but also to simplify management.
Cole: The way I understand it, your software extends network control all the way to the VM. What advantages does that offer, and does it help guard against provider lock-in?
Merchant: There are several advantages to extending network control down to the VM level. First off, it brings switching back into the network administrator's domain and out of the server administrator's domain. This results in more consistent application of network policies and ease of configuration and management. Secondly, the network can run at wire-speed without performance degradation. As such, leveraging the network instead of the virtual switch to deal with tasks such as access control and Quality of Service (QoS) works to free up the CPU cycles on the server and helps provide more consistent application performance. Lastly, leveraging the network allows support for heterogeneous environments where you may have multiple virtualization technologies in use.
As far as guarding against vendor lock-in, moving control back to the network goes a long way to help achieve that goal. The reality is that today there is a lack of standards around virtualization and the virtual switch in particular. Furthermore, the virtual switch model itself is in a state of transition. Hence it is very easy to be locked into a proprietary technology approach. On the other hand, networking technology is relatively mature and is agnostic concerning virtualization technology. So moving control to the network allows administrators to take a technology-agnostic view of virtualization and avoid vendor lock-in into what is really a technology in transition.