What Should Be Virtualized?
There’s several different reasons to virtualize a given server. What criteria should be considered when deciding whether or not a given server or application should be virtualized?
Consider your goals
Many IT organizations that I am familiar with have established management business objectives (MBOs) around virtualization. One popular MBO is to set a goal for 80% or more of servers to be virtualized. Such MBOs are usually fairly effective as they force an organization to be accountable for easily attainable metrics allowing for TCO benefits (power/space/cooling) to be more easily calculated and reported. Organizations that are receiving the biggest benefits from virtualization will typically have such an MBO and/or track their virtualization progress.
The motives for such an MBO may matter in the decision of what to virtualize. The organization may be focused on what I call first wave benefits of virtualization (consolidation, power/cooling, etc.). A good portion of the value proposition of the vBlock solution, is actually first-wave TCO reduction. Other motives for virtualization may be because all the other cool CIO’s are doing it, or they may be pursuing second wave of virtualization benefits from the private cloud model, which requires 100% virtualization. More on motives in a bit.
“Virtualize First” Methodology
A colleague of mine recently attended a webcast where the following slide was shown. I presume that this slide represents Dell’s virtualization methodology and is proposing it as a model for other organizations:
There’s a few things that I found notable on this slide. First of all, note the section on proprietary hardware – there are more options here as USB-pass-through is now supported in vSphere 4.1. If the required peripheral is USB-based you now have the option to consider virtualizing these servers with 4.1.
But the big thing I wanted to talk about was…..
In the workflow illustrated above, some decisions are referred to the virtualization team for further consideration. However, there appears to be an absolute rule that ALL “production database” servers not be virtualized, without recommending consultation with the virtualization team. What would be some reasons for this? Let’s take a closer look at the technical capabilities in this realm as well as motives and other considerations.
DATABASES: Technical Capability
There’s a stigma out there that OLTP and other demanding databases cannot be effectively virtualized. This was true for many cases a few years ago, but may no longer be true with vSphere 4.1 and Nehalem and later processors.
The VROOM! Blog at VMWare publishes a number of detailed studies of the performance of various databases. I’ll dive more deeply into some of these in future posts but here’s a quick summary of some VMware performance results that have been published:
- SAP at 93-95% of Native — 2500 SAP users (IBM DB2) on a single vSphere 4.1 host
- 171,600 “heavy” Microsoft SharePoint users on a single vSphere 4.1 host
- 8,000 Exchange 2010 users on a single vSphere 4.0 host
These tests demonstrate that a well-tuned virtual infrastructure is usually capable of running at more than 92% of native. This does not mean that a VM will only be 92% as fast as a physical server – rather it means that during peak loads during a stress test, there was a performance tax of up to 8% at it’s peak. Unless the server is running hot for significant periods of time, you may not even notice any significant degradation at all.
But that’s enough theory and whitepapers. What about the real world?
Virtualized Databases in the real world
EMC/VMware have virtualized their SAP and Oracle 11i apps and is actually a reference customer for Oracle. Check out published success stories for virtualizing SQL, SAP and Oracle. There’s the vBlock which large companies are considering for SAP and other mission-critical applications. And also consider this result from a recent Oracle User Group survey reported by The Register:
Sixty two per cent with Oracle on x86 use virtualization compared to 43 per cent for non x86-based Oracle sites. Cost savings associated with server hardware virtualization and an overall reduction in hardware costs were given as the leading reasons for companies adopting virtualization – 76 per cent and 59 per cent, respectively.
In the future I’ll post a more detailed article about virtualizing databases, but at this point is should be clear that mission-critical databases can be run successfully on vSphere. However since every datacenter is different one should consult with an organization’s virtualization team (and storage to verify IOPS), before making a decision.
Motives play a big role in whether or not virtualizing a database server is a good fit. If the motive is primarily the first-wave benefits of consolidation, the percieved benefits may not be sufficient for a risk-averse manager to pursue. But if the goal is to reap the benefits of the private cloud, or even just some second-wave benefits, then a strong case can often be made for virtualizing mission-critical ERP and OLTP systems.
The “Virtualize First” methodolgy as shown above is a good model, however I question the need to be so restrictive with database servers. Some managers have a perception of virtualization that is a few years behind current capabilities, so this may have been put in to appease such concerns and take a more cautious sales approach. If I could change one thing on the slide, it would be to refer database servers to the virtualization team. Databases certainly can be virtualized and there are significant benefits to doing so. Since every datacenter and database is unique, an internal analysis would be advised to determine if virtualization is plausible.