The Case for Live Migration
There are may use cases put forward for live workload migration and they are covered in the document, and were also covered in my previous blog “Making the Case for Long Distance Virtual Machine Mobility.” These use cases include optimizing server resource utilization, optimizing resource consumption at various locations, hybrid cloud where overflow workloads move to another data center, and disaster avoidance were workloads are moved to saftey, as well as a follow the sun model where workloads move according to the time zone of users. Some scenarios for data center migration for live workloads can be accomplished without implementing complicated first-hop-responder and route-optimization techniques. Other scenarios are possible however, they require implementing protocols that share state and fate, thereby increasing the complexity and reducing the reliability of the connected data centers. As a result of these considerations the network is a critical factor.
Network Design Considerations
The state of the network is critical to live migration since as the VM execution state is being copied, the physical hosts must both be connected to the same network and that network must have sufficient bandwidth to transfer this memory state and the latency must be minimal so that state synchronization can occur. Understanding design considerations and benchmarks, such as distance and latency numbers is critical. These are important since low latency is a requirement for clustering VMs because of the sensitivity of the heartbeat that detects a failure in the system and when stretching an L2 domain the geographical distance for interconnect data centers that is required to maintain reachability during live migration is constrained by latency. Points of congestion are another consideration in the data center architecture, and how to avoid bandwidth congestion at the MAN or WAN edge for the data center interconnect. Maintaining state is covered as state is necessary to keep sessions alive and avoid resets. Considerations for storage are also covered at storage must be reachable when workloads using it are moved.
Planning Ahead Leads to Success
When planning to migrate workloads it is critical to examine the business use case, because costs and planning requirements can be expensive and complicated. When designing your workload mobility solution, it is important to examine distance, and latency in your environment and to evaluate the dependencies that your server application flows have on each other and the network. Network administrators can achieve inter-data-center workload mobility by deploying a combination of VMware and Juniper Networks devices. VMware provides the mobility solution, and Juniper provides a high-performance, highly reliable data center LAN and data center WAN experience.
Juniper's Implementation Guide
To help customers understand how this technology works and to help them to deploy it more easily Juniper has created an implementation guide that provides design considerations and configuration details for server and virtualization administrators, security and network administrators, and designers as well as architects who want to extend their virtual machine resource clusters beyond a set of racks in the data center and migrate them over the WAN to other data centers. VMware vSphere, with its vMotion functionality, is the major virtualization product in the Enterprise x86 server market. Juniper has used this technology to verify the network design and implementation in our guide, however the concepts discussed in it can also be applied to other server virtualization products from IBM, Microsoft, Citrix, and others.
For more information see, Inter Data Center Workload Mobility with VMware Implementation Guide.