Thoughts on Virtual Infrastructure Management

Isn’t Virtualization Self-managing?

By Lisa Crewe

Server virtualization vendors have marketed their technologies as if they would automatically correct themselves and optimally “self load-balance” to assure great service levels to client applications. But infrastructure performance doesn’t depend on virtual servers alone. Domain-centric “dynamic resource” features fall short of providing the kind of service assurance that critical business applications require.

Total performance depends on end-to-end “cross-domain” resource capacities, configurations and competition (e.g. application, servers, storage, and network). If you’re unfamiliar with the term cross-domain, check out this recorded overview we’ve put together to explain it.

To truly achieve the promised ROI of virtualization requires the ability to actively manage the important performance and capacity opportunities provided. While hypervisor managers come with many “knobs and switches”, they don’t come with the enterprise knowledge or cross-domain visibility that would help optimally and dynamically set those knobs and switches. In fact, worse than not getting the full value out of virtualization is the risk that naive configurations and default policies might actually prove counter-productive.

Optimizing virtualized IT infrastructure requires a new type of analysis that accounts for not only cross-domain components and contention, but also models the impact of resource sharing “entitlement” settings, logical resource pool membership, dynamic re-assignment (migration policies) across clusters of physical resources, and non-linear system performance curves under shared workloads. These data center level metrics are also required to assure optimal performance of the virtual infrastructure before migrating to the private cloud.

Over the next several posts I’ll explain these data center level metrics in more detail and how you can use them to optimize while you virtualize.

November 18, 2009 | Filed Under Performance, Virtual infrastructure management | Leave a Comment 

How to Convince Your Boss You Need a Virtual Infrastructure Management Solution

By John Gavin

Gone are the days when simply telling your boss that you needed the coolest virtual infrastructure management product just like all the other cool companies doing production virtualization would get it for you . Let’s face it. That won’t work these days when layers of approval are common place, endless questions on capital purchases are the norm and budgets are tighter than ever with ROI and TCO part of every ones lexicon and reason for buying. Ok, so what do you do to get the boss into your camp? There are really three areas you should highlight to get your boss to say yes.

The first is an old standby – time to resolve problems. Your first line of defense is that the traditional element management tools you have for physically dedicated environments can be useful, but just don’t cut across all the devices and see the levels of virtual abstraction to give that integrated system view which is vital to understanding virtual environments. Those older tools aren’t designed to see through and track all the virtualized elements because they were designed to support one device type in a silo-like fashion. You know the endless hours that are wasted chasing data, going to meetings or sitting on conference calls trying to find the culprits involved in performance problems that are impacting critical application availability, could all be avoided. You suggest even a conservative assumption of reducing this wasted, always unplanned, activity by 50% would be worth the purchase alone. The average Akorri BalancePoint customer with 100 VMs can identify over $500,000 in staff productivity annually.
Read more

October 13, 2009 | Filed Under Customer Experiences, Performance, Virtual infrastructure management | Leave a Comment 

Clouds: The Big vs Managed fight all over again?

By Rob Strechay

In recent weeks I have become very enamored with the whole ferocity for which Clouds have become the “IT” thing to talk about. Be it a private cloud, one which a company builds itself, or a public cloud like the ones Amazon, HP, Google or Microsoft have built. DMTF has a group looking at cloud management and interop as well. In fact Network World had an interesting article saying that “only 15% of corporate customers have adopted or are considering adopting cloud technology over the next year”. Based on the companies I’ve talked to recently, this sounds about right.

I found another story really interesting about a disagreement between Google and Microsoft at the Structure 09 conference back in June. As recounted by The Register, Gill from Google talks about how they view the Cloud as horizontal versus Microsoft’s tuning the applications and Cloud to better service the applications, Google’s “approach is a little more absolute than [Microsoft's],” Gill said.

Gill went on to say, “Not only does getting to the end user have to be fast, but the back-end has to be extremely fast too…[We are] virtualizing the entire fabric so you get maximum utilization and speed on a global basis as opposed to local fixes – putting one service in a data center”.

He later used this example, “if we (Google) make a minor change to, say, disk storage to get a three per cent gain, and we roll that out to the GFS library, suddenly the entire base of applications stored on GFS sees that gain.”

So why recount this discussion from The Register article? Because I think the missing part from both of these approaches is that a Cloud has multiple applications vying for finite resources. You really have two choices. The first is to keep increasing the infrastructure capacity to keep up and exceed demand. The second is to plan to optimize the “top talker applications” to play more nicely.
Read more

July 23, 2009 | Filed Under Customer Experiences, Performance, Virtual infrastructure management | Leave a Comment 

The world does not revolve around VMWare!

By Rob Strechay

You know VMware is the 800 lb Gorilla in hypervisors, but there are others, citrix xen, IBM Lpars, Hp Npars, Sun Zones, and Microsoft Hyper-v (stop laughing this is serious).

I am going to let you in on a little secret. The Hypervisor is a commodity. What you are paying for are the software services on top of them. Yes you can argue some have more bells and whistles in the hypervisor, scale differently, support different OS’s better than others, or are cheaper. What you are paying for is the right to buy the software add-ons.

It will be interesting to see how each differentiates from each other over the next year. Who will say performance matters and deliver on service level guarantees? Which one will be the most cloud-ish? Will you care?

I think you will care, but I believe the one that builds the most complete eco-system will win. I believe it is like email wars; exchange vs notes. Most companies use exchange. Not because it has more bells and whistles or is the best “all-in-one” product. But because it is a very effective email and calendaring system. Where Lotus Notes was trying to be everything collaborative … MS gave you the options to pick off the Chinese food menu at a reasonable price.

Same will happen with hypervisors. You’re going to have the “all in one” vs the one with features core to the hypervisor (performance / stability) …

I would say the one place MS and others have never done well is management beyond fault and event. So who will be the differentiator in this space? Will it be build or buy? Who will care that not all the workloads will live in virtualized / cloud environments … many with the label of “business critical”?

I believe you will be looking once again for “best of breed” products. Products that were not meant to be panaceas or “single-panes-of-glass” but never got implemented. In my time in the IT industry I have seen people buy five or more single-panes and still never achieve any benefit. Like in the storage industry you will probably buy your element/fault management from the vender and get the rest from innovative vendors outside those vendors.

Editorial Note:
Recently some of the aforementioned hypervisor vendors have been saying they will provide you everything for your data center an even cook you breakfast in the morning. Most call this the “cloud” strategy. Clouds where you use the “services” … think web services … available. Problem is that you will not be able to simply write to one API can move workloads between “clouds”. For clouds, cloud suppliers, and such to be successful this will need to become reality. Supplying underlying infrastructure services with no easy way to write to them is only part of what would be a dramatic game changer. This is not to say “private clouds” in our IT shops will work, but leveraging “public cloud providers” will be difficult.

July 6, 2009 | Filed Under Akorri Blog Update, Customer Experiences, Performance, Virtual infrastructure management | 4 Comments 

Clouds or Fog: it is your choice!

By Rob Strechay

I recently gave a presentation on this Thornton May’s IT Value Studio. My feeling is that the concept of private vs public clouds, one virtualization flavor vs another, really will hamper the idea of cloud providers. Today’s cloud service providers are giving out virtual machines that are configured with particular application infrastructures. Not ideal for legacy application workloads. Current clouds are for “new applications” that are written to one or another clouds provider specs.

I also agree with John Gavin, CEO of Akorri when he said “Virtualization used to be all about the low-hanging fruit, where server consolidation and cost management were the drivers. Now it’s really all about simplifying management and managing performance issues when in mission-critical applications.” I agree with him, full disclouser, not becuase he signs my pay checks … but he is dead right.

But for cloud/grid/dynamic IT to become real, been saying this since my stint with GGF/OGF, not only do we need to stop recreating standards, but we need the right standards. One in particular that would nice to see adopted and would jump-start would be Open Virtualization Format (OVF) really being used. We need standards; maybe not even new ones but right ones.

Another issue that will have to be addressed is performance and service level management in this environment. Right now the number one reason people do not virtualize is the strength of business units. What I mean here is if a business unit within a company had a bad experience with virtual so they went back to Physical in some cases; not a death nail but the role out slows dramatically.

Virtualization is the new whipping boy for IT. This is why understanding (baselining) pre-virtualization like our customers do, then comparing and managing to “all systems are green” from an infrastructure perspective is so important. Mature virtualized organizations, some at 90%+ virtualized have realized this. When I am talking to them they rank stability, availability and performance as the top concerns they battle from the FUD of business units.

This is not rocket science. Baseline what you have (physical), convert and test (test baseline), make the decision to virtualize, measure impact to the production virtual environment, and compare ongoing operations versus the baseline. Like the directions on a shampoo bottle … shampoo, rinse, and repeat as desired.

June 23, 2009 | Filed Under Customer Experiences, Performance, Virtual infrastructure management | Leave a Comment 

Next Page »