|
1. Why should I virtualize inside a switch?
That is a very good question -- as I think that storage network vendors need
to provide a solid answer before you deploy this technology. The idea is to make it easier to deploy storage and provision it, and help make heterogeneous storage management more of a reality. The longer-range view is to provide more intelligent network management, ease provisioning of storage and start to automate a number of data management features such as snapshot, backup/restore and data replication. We still have a long way to go, and the market will not heat up until 2004 or even later depending on how customers like yourself react to this new approach. A key point is that virtualization is not a standalone product, but a foundation feature that is leveraged to do other management functions.
2. How do I achieve virtualization in a switch?
Virtualization will provided using a number of ways. Many storage networking vendors have not filled in many details on the technology they plan to use, but there are a number of common themes. One delivery approach is a dedicated switch that has ports that provide virtualization as a service to hosts connected into that switch. In general, this requires very intelligent ASIC technology that can communicate with both the array and host as well as handle the addiitional network traffic caused by the increased communication. This ASIC will also provide a lot of insight to management software about the flow of packets, availability, security and performance -- information that can be leveraged by management software on hosts and arrays. A second way, again using these intelligent ASICs, is to build a virtualization blade that slots into a storage director/larger fabric switch, which will provide virtualization via an onboard virtualization engine. In both cases, there is an assumption that you will start out only virtualizing part of your SAN, and build this deployment out
as you migrate to this new technology.
3. What issues might I come across when trying to achieve switch-based
virtualization?
We think there will be a number of issues to consider. Putting virtualization in the fabric will increase network latency, so performance
will be an issue if vendors do not appropriately plan for it in the specification process. Secondly, you need to consider redundancy. Does putting intelligence in the switch instead of arrays create a single point of failure if that switch goes down? You should ask vendors what their response is to this. And, lastly, how will this virtualization approach complement what you are already doing with management products and array management from other vendors. If it duplicates what you are already doing, you need to consider the need for this.
|
