Sunday, May 2, 2010

Cloud Computing - Notes from a guy behind the curtain

The latest buzz is Cloud computing. When I attended a CloudCamp hosted here in St. Louis, it became rather obvious that the term Cloud computing has an almost unlimited supply of definitions depending upon which Marketing Dweeb you talk to. It can range from everything from hosted MS Exchange to hosted VMs to applications and even hosted services. I'm really not in a position to say what Cloud is or isn't and in fact, I don't believe theres any way to win that argument. Cloud computing is a marketing perception that is rapidly morphing into whatever marketing types deem necessary to sell something. right or wrong - the spin doctors own the term and it is whatever they think will sell.

In my own perception, Cloud Computing is a process by which applications, services, and infrastructure are delivered to a customer in a rapid manner and empowers the customer to pay for what they use in small, finite increments. Cloud Computing incorporates a lot of technologies and process to make this happen. Technology like Virtualization, configuration management databases, hardware, and software.

What used to take days or weeks to deliver now takes minutes. MINUTES. What does this mean? It takes longer to secure an S Corp and setup a corresponding Tax ID than it does to setup and deliver a new companies web access to customers. And not just local, down home Main street customers but you are in business, competing at a GLOBAL LEVEL in MINUTES. And you pay as you go!

Sounds tasty, huh. Now heres a kink. How the heck do you manage this? I know the big Four Management companies say a relationship is more important than Best of Breed. I've heard in in numerous presentations and conversations. If you are in a position to sit still in business, this may be OK for you. Are you so secure in your market space that you do not fear competition to the point where you would sit idly?

Their products reflect this same lack of concern in that it is the same old stuff. It hasn't evolved much - it takes forever to get up and running and it takes months to be productive. For example, IBM Tivoli ITNM/IP. Takes at least a week of planning just to get ready to install it - IF you have hardware. Next, you need another week and consulting to get things cranking on a discovery for the first time. Takes weeks to integrate in your environment dealing with community string issues, network access, and even dealing with discovery elements.

Dealing with LDAP and network views is another nightmare altogether.

The UI is clunky, slow, and non-interactive. Way too slow to be used interactively as a diagnostic tool. At least you could work with NNM in earlier versions to get some sort of speed. (Well, with the Web UI, you had HP - slower than molasses - and then when you needed something that worked you bought Edge Technologies or Onion Peel.) In the ITNM/IP design, somebody in their infinite wisdom decided to store the map objects and map instances in a binary glob field in MySQL. At least if you had the coordinates you could FIX the topoviz maps or even display them in something a bit faster and more Web 2.0 - Like Flash / Flex. (Hardware is CHEAP!)

And how do you apply this product to a cloud infrastructure? If you can only discover once every few days, I think you're gonna miss a few customers setting up new infrastructures and not to mention any corresponding VMotion events that occur when things fail or load balance. How do you even discover and display the virtual network infrastructure with the real network infrastructure?

Even if you wanted to use it like TADDM, Tideway, or DDMi, the underlying database is not architected right. It doesn't allow you to map out the relationships between entities enough to make it viable. Even if you did a custom Discovery agent and plugged in NMap - (Hey! Everybody uses it!) you cannot fit the data correctly. And it isn't even close to the CIM schema.

And every time you want some additional functionality like performance data integration, its a new check and a new ball game. They sort of attempt to address this by enabling short term polling via the UI. Huge Fail. How do you look at data from yesterday? DOH!

ITNM/IP + Cloud == Shelfware.

If we are expected to respond at the speed of Cloud, there is a HUGE Pile of Compost consisting of management technology of the past that just isn't going to make it. These products take too much support, take too much resources to maintain, and they hold back innovation. The cost just doesn't justify the integration. Even the products we considered as untouchables. Many have been architected in a way that paints them in a corner. Once you evolve a tool kit into a solution, you have to take care not to close up integration capabilities along the way.

They take too long to install, take a huge level of effort to keep running, and the yearly maintenance costs can be rather daunting. The Cloud methodology kind of changes the rules a bit. In the cloud, its SaaS. You sign up for management. You pay for what you get. If you don't like it or want something else, presto changeo - an hour later, you're on a new plan! AND you pay as you go. No more HUGE budget outlays, planning, and negotiation cycles. No more "True-Ups" at the end of the year that kill your upward mobility and career.

BMC - Bring More Cash
EMC - EXtreme Monetary Concerns
IBM - I've Been Mugged!
HP - Huge PriceTag!
CA - Cash Advanced!

Think about Remedy. Huge Cash outlay up front. Takes a long time to get up and running. Takes even longer to get into production. Hard to change over time. And everything custom becomes an ordeal at upgrade time.

They are counting on you to not be able to move. That you have become so political and process bound, you couldn't replace it if you wanted to. In fact, in the late 80s and early 90s, there was the notion that the applications that ran on mainframes could never be moved off of those huge platforms. I remember working on the transition from Space Station Freedom to International Space Station Information systems. The old MacDac folks kept telling up there was no way we could move to open systems. Especially on time and under budget. 9 months later, 2 ES9000 Duals and a bunch of Vaxes repurposed. 28 Applications migrated. Reduced support head count from over 300 to 70. And it was done with less than half of the cost of software maintenance for a year. New software costs ~15% of what they were before. And we had alot more users. And it was faster too! NASA. ESA. CSA. NASDA. RSA. All customers.

Bert Beals, Mark Spooner, and Tim Forrester are among my list of folks that had a profound effect on my career in that they taught me through example to keep it simple and that NOTHING is impossible. And to keep asking "And then what?"

And while not every app fits in a VM, there is a growing catalog of appliance based applications that make total sense. You get to optimize the hardware according to the application and its data. That first couple of months of planning, sizing, and procurement - DONE.

And some apps thrive on the Cloud virtualization. If you need a data warehouse or are looking to make sense of your data footprint, check out Greenplum. Distributed Database BASED on VMs! You plug in resources as VMs as you grow and change!

And the line between the network and the systems and the applications and the users - disappearing quickly. Presents an ever increasing data challenge to be able to discover and use all these relationships to deliver better services to customers.

Cloud Computing is bringing that revolution and reinvention cycle back into focus in the IT industry. It is a culling event as it will cull out the non-producers and change the customer engagement rules. Best of Breed is back! And with a Vengeance!

No comments:

Post a Comment