SearchServerVirtualization.com: What are some hurdles current ESX administrators might face with during the upgrade...
Dr. Stephen Herrod: As with any major product upgrade, especially a mission-critical one, the most important element is to plan the upgrade.
There is a tremendous amount of new functionality in the products. We've put several things in place to help encourage and manage the transition. A few specific areas are worth calling out:
- - Virtual Center: Virtual Center 2.0 can manage mixed ESX Server version 2 and 3 environments. This allows you to take advantage of the new management functionality while transitioning your ESX servers.
- - Virtual Mmachines: Virtual machines (VMs) created on ESX Server 2.x can be brought into the new ESX Server 3.0 environment and will run fine without modification. In order to take advantage of the latest virtual hardware and features like four-way SMP, a virtual machine does need to be upgraded.
This is a one-way path upgrade, so customers will want to clone their virtual machines beforehand if they think they'll want to continue to use them on ESX Server 2.x hosts as well. For mass upgrading of a large number of the VMs (virtual machines) from Version 2 to Version 3 (that is, upgrading the virtual hardware, the on-disk file format, and the VMware Ttools), we've provided a batch VM upgrade script.
- - Virtual Machine File System (VMFS): VMware ESX Server 2.x uses version 2 of our VMFS file system. This file system should be upgraded to the newest version in order to take advantage of the VMware Infrastructure features.
We have several tools to aid in this move. For example, we support in-place upgrading of an existing VMFS 2 file system. This allows you to upgrade without any additional temporary storage needs. We also allow what we call "cold migration" to copy VMs from a previous VMFS 2 file system to a new VMFS 3 file system. Both processes do require some virtual machine downtime, so this needs to be planned. To avoid this downtime, we are actively working on a method of seamlessly moving live VMs between versions.
Has VMware considered forming a partnership with the makers of Splunk to enable a more dynamic community analysis of ESX logs?
Herrod: It's a good idea and one worth exploration. I do know that some of our customers use splunk for analysis, and I also know that several customers are running the splunk software itself from within VMware virtual machines.
Microsoft believes that virtualization should be a component of the operating system, and XenSource believes that it should be a component of the infrastructure. Where does VMware believe that virtualization falls in the scheme of things?
Herrod: We believe that virtualization should be thought of along the lines of hardware. More specifically, we find that many of our customers treat it as a part of their compute and storage layer, thinking about it at hardware purchase time and expecting the same levels of reliability and performance. We also believe virtualization services should be independent of the operating system running within virtual machines.
Tight integration comes at the unfortunate cost of giving up a bias-free choice of operating system and thus the software stack (i.e., OS and application program). Customers who already have mixed software environments or who would like to have that option in the future will benefit from this lack of OS bias. And finally, tying the virtualization layer to an OS could push towards additional software dependencies.
Should you have to upgrade your virtualization layer if the guest operating system requires patching? We don't think so.
When is virtualization going to facilitate high-level grid computing?
Herrod: Grid computing is an interesting topic that often means different things to different people. I think about grid computing as having a few salient features: a computational workload broken down into components that can be flexibly scheduled across a pool of hardware resources, a workload that benefits from the availability and cumulative throughput that comes from these dynamic and flexible mappings, and a collection of hardware that can be more highly utilized via load-balancing of multiple grid-using applications.
With that definition, I believe that VMware Infrastructure is already delivering upon many of the most desirable features of grid computing. Because virtual machines are fully encapsulated and hardware independent, they can run unmodified on a diverse set of hardware. Secondly, the combination of resource pools and VMware DRS [Distributed Resource Scheduler] allows these virtual machines to be flexibly and dynamically scheduled across your physical hardware. Thirdly, we bring many of the availability benefits of grid computing via technologies such as VMware HA. VMware HA recognizes if a physical server goes down and automatically restarts the virtual machines that were running on it on other available physical servers. And finally, virtualization has always helped increase the utilization of hardware by consolidation as well as dynamic load balancing.
Could you elaborate on the utilization issues?
Herrod: The resource pool abstraction is meant for different services, and we believe that a grid compute cluster is often just one client of the IT infrastructure. If customers want to provide only 10% of resources to the Ggrid Ccluster in the day and 50% of resources at night, they should be possibleable to do that, and we can do exactly thisat with our schedulers.
Different services can exist in different resource pools in any organization. We can optimize the resource utilization process by either saving power (i.e., by using intelligent algorithms to move VMs around and powering off servers). Another direction is to use the underutilized servers for grid computing workloads.
Grid computing is as multilayered as an onion, of course.
Herrod: There are many other components to grid computing. For example, we do not currently ship a job submission front-end. Similarly, we don't currently provide some of the accounting and reporting capabilities of today's grid solutions. These job management capabilities are often implemented by vendors such as Sun, Platform Computing, United Devices, Oracle or IBM and could be built using our Virtual Infrastructure software development kit (SDK).
Overall I think we are making progress in the grid computing space but are not truly packaged as a grid computing solution today. Moving forward, we'll work both internally and with our partners to deliver even more packaged capabilities towards full grid computing solutions.If you would like to read more from Dr. Stephen Herrod, please check out Virtually There: Steve Herrod's Blog.