Monday, July 12, 2010

George Crump, Senior Analyst

As server virtualization took hold in the data center, live migration of virtual machines from host to host was looked at as an availability tool. As the technology matured organizations began to leverage features like VMware's VMotion for more than just availability, organizations began to use the non-disruptive migration capabilities to load balance processing resources throughout the organization. Live migration is a significant move forward in creating a utilization balance within the data center. The problem with live migration is it is still counting on spare resources being made available somewhere in the data center. This use case still impacts the data center budget negatively since dollars must be spent procuring and maintaining potentially underutilized resources.

Cloud bursting attempts to change this by leveraging VMotion and enhancing it with SAN extension products like Brocade's 7800 SAN Extension Switch or Brocade FX8-24 Extension Blade to provide live migration between an organization’s data centers or even to a cloud provider’s data center. When correctly implemented VMotion over distance provides data center managers the ability to shift an application’s virtual machine to another data center in real time, not interrupting user workflow. It smooths out spikes in demand.

In essence the organization practices the purest form of cloud computing where regional data centers are no longer walled off from each other but actively participate with each other more efficiently using idle compute across multiple data centers. In short the IT team can respond to the needs of the organization as quickly as those needs surface or change. Most importantly cloud bursting with distance VMotion has the potential to significantly drive down cost by not requiring each data center to build for the peak use case but instead build for a small percentage more than the normal workload.

The Challenges to VMotion over Distance

When looking to extend VMotion geographically one of the issues to contend with is latency. VMotion has a relatively short latency requirement of typically 5 milliseconds. That means that the maximum distance it can support is less than 200 miles.

Additionally, for everything to move seamlessly, the entire infrastructure will need to be on the same layer 2 subnet. Layer 2 tunneling across distance can be done via Virtual Private LAN Services (VPLS) using Multiple Protocol Label Switching (MPLS) for tunneling Ethernet frames over the IP network. The VPLS allows IP addresses and storage routing to stay fixed; another requirement of VMotion. As most customers deploy connectivity between data centers relying on service provider networks, and MPLS is a well know and a commonly deployed service provider protocol, it’s a good building block for extending the L2 Ethernet network between regional data centers.

An important decision is how to ensure the data can still be accessed. The two choices are either across the wide area connection or to make it available locally. If a WAN connection to the data is used, latency goes up and copying the data between data centers when a VMotion is performed requires a Storage VMotion which can take a long time. That’s why synchronous remote data replication (RDR) is often preferred so consistent data is available at both locations. The most common networking protocol supporting RDR is Fibre Channel over IP (FCIP). FCIP is a storage tunneling protocol that enables IT organizations to move more data faster and further than ever before over lower-cost, shared IP WAN networks. These are capabilities that can either be added to existing backbone infrastructure products like Brocade’s DCX Backbone with the FX8-24 Extension Blade or via adding Brocade's 7800 Extension Switch. Both ensure synchronous storage replication traffic on shared IP WAN segments is not disrupted by other IP traffic flows. FCIP enables the SAN to ensure local data access to the remotely started VM when needed. These types of SAN extension products provide Storage Optimized TCP (SO-TCP), Adaptive Rate Limiting, and storage protocol optimization to increase the distance between data centers and minimize effective latency while ensuring replication traffic has the dedicated bandwidth it requires (typically OC-12 or higher) when sharing WAN links with other traffic.

Drive Down Cost, Increase Agility

Enabling cloud bursting with Long Distance VMotion allows the IT department to further drive down cost by optimizing the resource utilization within each data center. This can be done without risk of cutting resource allocation too close when above normal resource requirements occur. With VMotion over distance, VMs can be balanced across multiple regional data centers as needed with no service interruption to the users of those VMs. This risk can be further mitigated by leveraging a trusted cloud compute provider that can host VMs if other data centers can’t manage the increase workload. The combination allows the IT department to provide the organization with unmatched flexibility without breaking the budget.