On Monday Oracle announced their new Xen-based hypervisor - Oracle VM (OVM). Oracle's timing for entrance into the virtualization space is excellent, considering that they are leveraging the Xen hypervisor which is nearing full maturity, and since the majority of enterprises have plans to virtualize production servers in 2008. Oracle is touting some very impressive performance numbers, but have yet to publicize them. Of note, however, is that Oracle verified performance of their hypervisor by running four concurrent VMs on a physical host. Too many benchmarks reflect performance of one or two VMs running on a physical host, which is not a very accurate portrayal of most virtualized environments. So to Oracle's credit, they are using realistic scenarios in their benchmarks. Thus far Oracle has said that they are seeing 3X greater efficiency of Oracle products running on Oracle VM than on existing x86 virtualization platforms (read: VMware). Naturally, I would like to see a published benchmark confirming these results. A 3X bump in performance is a pretty strong claim, and Oracle should publicly disclose their benchmark test case so that other independent bodies can validate their claims. Until that time, I would suggest that all organizations conduct their own performance evaluations and draw their own conclusions. Note that Oracle is also officially on record as saying that "Oracle VM is the only x86-based server virtualization environment on which Oracle products are supported" (Source: OVM FAQ).
OK - now that the word is out, let's take a look at what this all means, starting with the good news first. Oracle VM represents another hypervisor choice and will certainly be looked at with interest by existing Oracle customers. The reason for that? Oracle customers have a one stop support shop for their Oracle applications, OSs (assuming Oracle Enterprise Linux), and x86 virtualization platform (OVM). And as Bob Barker would say "The price is right." Oracle VM support subscriptions are priced at $499 per year for 2 processor systems or $1,497 per physical server instance (unlimited processors). Oracle is working on paravirtualized device drivers for other supported operating systems (including Windows Server 2003) and expects the drivers to be available by early 2008. OVM includes a pretty nice management tool, making centralized VM monitoring and deployment a simple process. Oracle has also committed to the DMTF's open virtual machine format (OVF), and plans to support OVF in 2008 once the OVF standard is finalized. I asked about Oracle's intention to support the DMTF's CIM management profiles for virtualization and have yet to receive confirmation.
Oracle's refusal to fully support other x86 virtualization platforms (most notably VMware ESX Server) hints strongly at Oracle's market penetration strategy. While Oracle does not publicly support VMware ESX Server, it should be noted that Oracle has historically supported their applications on ESX platforms on a case-by-case basis with their larger enterprise customers. Now back to the support issue. Full support for Oracle applications on OVM does give Oracle an "in" with enterprise customers. I'm sure that their strategy is to first get OVM into enterprise data centers and then work to chip away market leader VMware. Still, it's hard to find a solid argument as to why Oracle does not publicly support their applications on ESX Server. Based on Oracle's performance assessment, it looks like Oracle plans to stand behind performance "concerns" as the reason for sticking with only supporting their applications on OVM. I'd like to see Oracle's benchmark test case and validate it in my lab before agreeing with Oracle's performance assertions. I have worked with numerous clients that have successfully run Oracle databases in VMware VMs for over a year without incident, so I feel that Oracle would do the industry a service by officially adding support for the ESX hypervisor. VMware was quick to counter Oracle's performance assertions and posted a blog response of their own.
Many x86 virtualization vendors see 2008 as the year they begin their assault on VMware. Their rally cry will not initially be one of superior features, but rather one of "we're good enough" combined with excellent management tools. Feature-for-feature, no one is ready to compete with the ESX hypervisor, which so far stands to be the lone hypervisor that will not be commoditized (there's far too many unique features to the ESX hypervisor to call it a commodity).
Now let's look at what equals "good enough." In my opinion, the following features are essential in x86 virtualization deployments (features currently present in OVM are marked with an asterisk):
- Live migration*
- High availability (dynamic VM failover or restart triggered by a system, OS, or application failure)
- OS vendor (OSV) certification and support
- Independent software vendor (ISV) certification and support
- Independent hardware vendor (IHV) certification and support
- Backup/data protection vendor certification and support
- Centralized management
The above features, to me, are the deal breakers in virtualization platform selection used for production systems. Some organizations would like features such as VMware's distributed resource scheduler (DRS), but find that they can live without it if need be. Virtualization platforms that do not offer features such as mature high availability should not even be part of the virtualization discussion until they do. Also, most organizations don't like to mix the words "backup" and "unsupported" in the same sentence. Oracle's intention to support Windows Server 2003 is noteworthy; however, Microsoft has yet to confirm that they will officially support OVM as a virtualization platform. Currently, you should expect "best effort" support. I have inquired with Oracle regarding if and how they will add high availability to OVM and am awaiting a response. So for now, I'm making the assumption that an OVM HA solution will appear sometime in 2008. Until that time, I see OVM as a potential match for development and test environments, but not yet ready for use in production. In addition to HA support, I'd need to see OVM fill out my checklist above before recommending it as a production x86 virtualization platform.
Now that Oracle is in the x86 virtualization space, it's also time for them to slightly revise their licensing policies to be more favorable to x86 VM environments. VM portability is much easier to manage when product licenses are not tied to physical hardware (which may be different as a VM relocates to new systems to facilitate scheduled or unscheduled maintenance and dynamic workload balancing). Nearly every major enterprise software vendor is at least considering offering customers a virtual CPU-based licensing or support alternative. While Oracle should keep its physical support subscription model (ideal for physical systems, LPARs, etc.), it's time for them to consider a support subscription model that allows VMs with support subscriptions to move between physical systems without restriction or penalty. For example, two product instances on one physical machine requires one license. However, if one of those instances is moved to another machine you need a new license. Even a license transfer can be complex if a target system has a different physical CPU count (i.e. 2-way to 4-way) than the source system.
Overall, I have mixed emotions about Oracle's entrance into the virtualization space. Oracle has had limited success with Oracle Enterprise Linux. But I can see organizations interested in the idea of Oracle VM appliances running OEL, with OS updates, application updates, and support all coming from one source. That being said, Oracle applications run on numerous operating systems today and I worry that a proprietary support model could eventually backfire. Oracle's ownership of the application and virtualization stack will be appealing to some, but organizations with existing VMware ESX infrastructures are not going to be in a hurry to replace their ESX servers (with established maturity in the enterprise) with OVM. So if Oracle is using performance to justify not publicly supporting VMware ESX Server, they need to show us the evidence. Your honor, where is exhibit A?
I applaud Oracle's entrance into the virtualization space and in general see added competition as good for the industry overall. However, I'd prefer to see Oracle win on merit. Oracle has a talented bunch of engineers and shouldn't be afraid to go head-to-head with VMware, Microsoft, and Citrix. Let's not use support restrictions as a means to strong arm customers toward OVM. Instead, show us a great virtualization platform, and the customers will come on their own.
Posted by: Chris Wolf
Comments