VMware's latest salvo in its virtualization war with Microsoft is vSphere 5.5, which features a host of improvements, the most interesting being high availability, support for Big Data/Hadoop and improved storage and backup.
The ongoing battle pits Microsoft's Hyper-V V3 hypervisor and the Microsoft System Center 2013 management platform, now with R2 updates, against vSphere 5.5 and vSphere Operations Manager (vSOM 5.5).
Some of the vSphere/Operations Manager bits are poised towards infrastructure, like those that handily unite vSAN storage options or can use server-based flash storage. But the best and most interesting bits in the release are high availability features within specific application structures. This traces back to VMware's acquisition of Hyperic. One note of caution -- high availability needs to be important to you, because it's only found in VMware's most expensive option.
In our testing, we found that the other vSOM5.5 highly touted features -- VM movement, cloning, better and denser vCenter UI, and snapshot restoration capabilities -- worked largely as expected. There is also support for extensions to clusters of VMs of the type that support distributed databases, especially Hadoop infrastructures. Unfortunately, there are few standards-based methods to test Hadoop and Big Data.
We reviewed Hyperic's VM management app five years ago. Its new role is as an application monitor and re-starter in a formula called App HA. We also found that the products covered by this formula are a handful of mostly Microsoft apps, although Apache and Tomcat on Windows or Linux can be covered, too.
It wouldn't take much to monitor currently non-supported apps. And despite a fit of early documentation oddities, lack of unifying syslog management to ease troubleshooting, and a few other small scrapes, we made it all work.
So, should you run right out and upgrade to vSphere 5.5? The answer is yes, if maximized app uptime for Microsoft-specific or top-tier database and web apps can be a job-changer. And if you have the budget (or have a well-negotiated contract), as most of the new features come at the top tier, per-CPU pricing.
Not much in this release is poised towards cloud, private or public, that hasn't been already introduced. Your Adobe Flash must be upgraded, too.
The feature list for VMware vSphere 5.5 with Operations Manager is graduated in terms of Standard, Enterprise, and Enterprise+. Only Enterprise+, at $4,245 per CPU, has the juiciest features of App HA, Big Data Extensions (Hadoop support), and hybrid flash-cache read support.
As Microsoft has aimed Windows 2012 R2 and Systems Center 2012 R2 at the exposed throat of VMware, the HA changes are the secret sauce in 5.5, along with a maturation of Active Directory Single Sign-On (SSO) initiatives.
Here, VMware is closer to neck-and-neck with Microsoft's support for their own products, as VMware attempts to compete with Microsoft's vast and pending upgrades to Hyper-V virtualization features and Microsoft System Center updates.
VMware 5.5 does focus on topics other than lobbing competitive grenades at Microsoft, such as the ability to run more and larger Big Data constructs, and releases extensions for Hadoop clusters. Hadoop infrastructure popularity seems to grow out of control as it takes a front seat in the database segment once labeled "business intelligence/BI." VMware's clustering components are an answer to the mushrooming of Hadoop clusters that might be brought up on other competitor's turf -- and the target in this case is painted on Red Hat's back.
Install and update
As this is a simple dot-upgrade, not much should have changed, and we were able to install ESXi 5.5 in the lab and at our network operations center at nFrame/Expedient with staggering ease. Fresh installs appear to have more drivers available than earlier versions, although it's still best to check with VMware's hardware compatibility list first. We found a few important drivers to match our hardware that were missing in vSphere 5.1.
Preparation also needs to be made to protect applications transitioned into App HA. Although the base supported operating systems and supported applications are designed to be protected, this doesn't mean application logic doesn't need to be examined and addressed. The middleware and external application logic that surrounds most production database activities aren't necessarily covered by App HA actions.
As an example, if a database is alive and ticking, but uses a "non-standard" listening service/RPC, the service needs to be able to also login after a restart externally (as in triggered by its own logic) to, say, SQLServer or the Windows operating system itself. This means that applications that monitor SQL Server, Oracle, or a Windows 2008R2 or 2012 or Windows 7/8x instance must be able to reconnect after a time-out/restart cycle that App HA has managed. Then other apps must be smart enough to reconnect and synchronize. In some instances, it's not as easy as it looks.
App HA plugs initially into the vCenter appliance. The way that App HA Appliance works is to watch the Windows Service for an unavailability. If a VM (operating system) totally fails, the reaction can be set by policy to restart it. We installed an appliance, which has Postgre for a database and a Java-based engine, and tried both SQL Server 2013 on a Windows 2008 R2 VM as well as a generic Apache installation on Linux Centos -- or Oracle Linux (not tested). The method and approach to each is similar, where a process monitors each desired application instance autonomously through an agent. Conditions are imposed on what constitutes the need for action, then what action is taken.
The App HA worked for our two tested apps and VMs. Kill them abruptly, and they're resurrected by command, like zombies. In our small tight network, the results were essentially instantaneous. OK, they were as healthy as any other restarted application that went down "hard" and therein lies a task list for automated recovery on the part of IT staff where App HA is deployed and running. Transactions in progress, connectivity drills, and other restart efforts must be addressed.
There's a Hyperic-ish App HA virtual appliance that must be in communication with the monitoring process in covered VMs for things to work. Communications failures will be catastrophic, but this would be true without App HA. It's up to other application logic to perform tasks like transaction tracking and restoration, re-authenticate to other data sources, or otherwise keep things in synch, and so App HA can be built-in, but doesn't cover every base of say, all systems app logic of a SQL Server-based accounting application. New projects will be able to easily incorporate App HA thinking, but retrofits may require mercurial re-coding tasks and test prior to placement into production.
Improved bit camera
More information is now stored with snapshots, and versioning of data backups/snapshots is easier than ever, thus making backtracking VMs or even storing trees of snapshots comparatively simple -- if you have the disk space.
Backup is easier than ever, is included in a basic form, and doesn't require some of the comparative pieces that are optional in competitive platforms. We found it simpler to use. The vSphere Data Protection comes as an appliance in two forms, one with strong base features, and a second that's more easily scalable, if at an extra price. No host client app is required, as it's all controlled through the vCenter client, or web user interface at the vCenter Appliance. There are options to more gracefully save state for Microsoft Exchange and SQL Server and we didn't extensively test these options.
The score so far is two appliance-based feature additions, and this is a trend. The core appliance we used was uprated from prior vCenter versions, and the vCenter UI worked very well most of the time. Occasionally, we had greyed-out choices for no reason we could understand. Usually, climbing back a few logical web pages could cure the problem and options would re-appear. The UI mandates Adobe Flash 11.3+, which limits choices on older operating systems versions that Adobe no longer supports, and the list was longer than we had imagined. Perhaps Adobe's security has improved.
The vCenter web UI changes the nexus for control further away from the vCenter Windows client-based application, which was automatically updated from prior versions when we tried to manage a vSphere 5.5 host. The vCenter Windows app can no longer be used on Windows domain controllers at all, and so with only Windows client compatibility to run on, the vCenter UI becomes perhaps the future singular nexus for vCenter operations. In as much as the UI has improved dramatically from its first iterations a couple of years ago, we applaud the move.
VMware also has added extensions for Hadoop, although we didn't test this, lacking a useful standards-based method to do so. We also were unable to test vSOM5.5's fast-read capability based on flash-cache. A flash drive pool can be allocated for faster VM reads, VMware says, adding to overall application performance. Support for PCIe bus Flash SSDs also includes hot-pluggable drives, which given our higher-than-conventional-drive failure rate for SSDs, is a potentially good feature.
For those running VDI, support has been added for advanced AMD and nVidia GPUs. VMware vMotion now also allows VM movement from one advanced GPU family host to another without explosive results.
Concerns are small
There is a lot of vProduct whose vFunctions are eVer more difficult to compare and contrast to those of vCompetitors. The places where the various VMware products and feature sets connect to each other beg for a homogenization and stratification of functionality and the documents that would glue things together. The feature sets often have the feel of Velcro, rather than the strength of turbine-propelled aircraft aluminum that has been the traditional feel of VMware. Although VMware is a pioneer, components that build varying systems become increasingly opaque to us as the product lines expand, acquisitions are amalgamated, and iterative releases bear new feature sets.
Iterative releases can also complicate relationships among hosts, host services, VMs, and other infrastructure at both IPv4 and IPv6 levels. Fortunately, VMware's almost maniacal adherence to Microsoft Active Directory best practices means there is much authentication among RPCs that is entwined with Microsoft Active Directory authentication policy.
Although an incremental release, those with investments in VMware's infrastructure will want to take a thorough look at vSphere 5.5. Whenever 6.0 arrives, VMware is pointing towards more of web-based, rather than app-based, control of virtualization infrastructure with a trend towards software defined networking.
Also, look for extensive management controls over almost total process virtualization. The App HA components are an interesting start, and APIs that would allow broader cross-platform application availability may be in the offing.
Henderson is principal researcher for ExtremeLabs, of Bloomington, Ind. He can be reached at email@example.com.
Read more about data center in Network World's Data Center section.