Archive for May, 2011
First, my definition of a cloud, not that it’s any better than others, just my vantage point so you know where I’m grounded. Then some insight on how I’ve seen the evolution of cloud systems, and how the industry arrived at what we have today. Finally, a summary of current vendor offerings and what they can do for executing the enterprise cloud strategy. Really the offerings are staggering, so I’ll break this topic up, with this post touching on the private cloud infrastructure plays.
What’s a cloud?
The current nomenclature being used today consists of Infrastructure (IaaS), Platform (PaaS), and Software (SaaS) and defines the boundaries for the different types of cloud services:
Infrastructure clouds provide virtualization of severs and operating system to optimize server utilization, and simplify operating system, storage, and network provisioning, along with elastic runtime behaviors. A public example of this is Amazon Web Services Cloud < http://aws.amazon.com/ >. With the appropriate monitoring in place this approach helps significantly with capacity planning. As at some point there is always real hardware to deal with, and that never goes away.
Platform clouds are an entire framework to build and deploy applications, reducing time and complexity for software construction as well as deployment, while also abstracting out all the infrastructure components.
Public examples of general platforms are:
|Google App Engine||http://code.google.com/appengine/|
Yes, salesforce.com is moving from just a web based CRM solution to a complete platform.
Software clouds are entire applications for any number of globally distributed end users, typically utilizing some PaaS for its implementation. For example, Salesforce.com has Sales Cloud < http://www.salesforce.com/crm/sales-force-automation/> that uses Force.com, and Oracle CRM On-Demand < http://crmondemand.oracle.com/en/index.htm >. For Social media software services, Facebook, and provides Facebook Apps as a platform < http://developers.facebook.com >.
All clouds provide elastic resources on-demand that ultimately maximizes the quality of service, all for a price point lower than the custom build-out and support for each individual application, platform, or infrastructure component. Clouds also provide the metering and administration required to effectively implement a billing or charge back model, providing an on-demand pay for what you use model to the service consumer. So that is how I view the current cloud landscape. Although I gave public facing examples, an enterprise cloud strategy will be a combination of public, private, and joint venture services.
What brought us here?
Actually cloud computing is little more than the mainframe model that has been around for decades. Since System Z obtained the ability to spin up Linux LPARs, run application servers, relational databases, and provides MQ or RPC-like connectivity into CICS and IMS, it’s evolved into a fairly open cloud system. That’s why the mainframe folks wonder what people are talking about when we talk about “the cloud”, since they have been living it for decades, and 10 years of that has been an open compute model. So why did it never take-off to eliminate the mid-range server sprawl of the data center? Simply, it was not lack of technology, mostly just cost for hardware and software. Some of the pricing justified, but much of it not, and this is a whole other topic. However, as a technical model the mainframe is already a cloud system, so why is cloud computing being heralded as such a new achievement? Well it’s not a revolution, but an evolution and some great marketing rap.
As of 10 years ago the mainframe model became an extremely open virtualized platform, although its inherit cost prohibitive nature really beat down innovation. Meanwhile, virtualization on mid-range computer and storage systems at a reasonable price tag garnered the attention of the masses, increasing utilization and manageability, thus optimizing the cost model. A continued quest to reduce the complexity involved in environment management and provisioning of the software services that use them was underway. Simultaneously, the application service provider model was moving to a full platform as a service (PaaS) approach, again at reasonable prices, increasing its popularity. Vendors such as Google and Salesforce drove SaaS and PaaS offerings, Amazon brought infrastructure as a service to the public, and finally Microsoft with PaaS Azure. All those efforts collided to ignite current day cloud computing giving something for the marketing teams to rally around, attracting dollars to fuel further innovation, accomplishing two important things. It brought the mainframe model, with most of the performance to a reasonable price, providing it over a wide range of existing vendor hardware and software. That is where the innovation rest, not in a revolutionary new way of doing things that no one ever thought of before (the hype), but evolving what we know into something usable by more people, and provided by more vendors to continue to fuel the innovation. However, don’t declare the mainframe a dead platform (again) it’s evolving as well. I really hope we revolutionize the way we manage technology, but for now, it’s an evolutionary grind.
What follows is a brief summary of the IaaS landscape and my thoughts on where that general approach is headed.
IaaS Vendors: Oracle, IBM, HP, and Cisco
The offerings are all very similar following the model of blades in a rack with storage all connected via 10Gb Ethernet or 40Gb Infiniband. Then connect multiple racks together to scale horizontally. Tied all together with virtualization and provisioning management. OK I oversimplified it, but in short, they are all starting to look like the mainframe model. Now the with IBM’s z196 out that has a zBX chassis to support Xeon and Power blades under mainframe management, the lines are really starting to blur.
Below is a very brief summary of each vendor, actually the offerings are impressive, and sorting through them all is time consuming. Mostly my question after reviewing these cloud offerings is how do we keep from creating silos of cloud infrastructure? Meaning Oracle cannot be managed by IBM, which cannot be managed by Cisco. It’s really not any worse than it is today, but it’s not any better, and is concerning since cloud computing was supposed to virtualize and optimize our environments. We get those benefits, but with boundaries, and one has to wonder just how much benefit is really being achieved by the “cloud in a box” mainframe model solution. I think it’s better to have borderless cloud computing meaning virtualization and software provisioning that lays across all hardware platforms, provides high-speed interconnects, and virtualizes the IP network across data centers. Of course the midrange solutions are cheaper than mainframe, and their performance is approaching the mainframe, but at what cost to real manageability across the entire enterprise? To me that is the real challenge that information technology faces, and is not solved with a cloud in a box, but that approach is a step on the right path.
Oracle Exadata and Exalogic
Oracle ties together storage, database, and application services with high-speed low-latency InfiniBand in a rack. Exadata is a database appliance, and Exalogic is an application server appliance, however, Oracle does not limit the software that can be deployed and managed within their Exalogic platform, so it can be also viewed as a general cloud infrastructure, or as Oracle describes it, a cloud in a box. All operating systems are virtualized, and then software provisioned on the virtualized environments, with multiple Exalogic racks being hooked together. Interconnect is 40Gb Infiniband. The Exadata platform is more of OLTP configuration, and is getting some heat brought to it when compared to TerraData or Netezza. Not sure I would do a direct comparison, Oracle’s strength is OLTP, and TerraData was designed for DW/Analytics.
Exalogic: The One Day Installation Challenge
HP Cloud System
HP is marketed as general purpose cloud platform using HP hardware (network, storage, servers), in addition to being able to provision onto public clouds, such as Amazon’s EC2, or HP’s own cloud service. With the blade servers and Infiniband interconnects, it looks very similar to the Oracle solution. Although, HP does not require total virtualization for the provisioning and management like Oracle does. Meaning it supports OS provisioning without a hypervisor. All of these features seem to move in more of the direction that is required for generalized cloud platforms.
Here is an informative article on HPs offering, and I think describes HP approach very well http://jshurwitz.wordpress.com/2011/02/15/hps-ambitious-cloud-computing-strategy-can-hp-emerging-as-a-power/ .
Their IaaS components are similar to HP and Oracle; with Cloud Burst on Xeon, Cloud Burst on Power Systems, and their zSystem cloud for the mainframe. Their latest z196 Mainframe integrates a blade chassis (zBX) over 10Gb Ethernet, to utilize their blades like DataPower, Xeon, or Power, then bring those blades under mainframe management. How all these cloud platforms can actually integrate in terms of providing a seamless cloud fabric is unclear, but if possible would be compelling to those enterprises that have blue coursing through their technology veins.
Cisco Unified Computing System
Similar to Oracle, HP, and IBM; uses Cisco servers and network. Cicso servers are using 10Gb Ethernet and FCoE for the interconnect. Obviously they have more advanced network virtualization and provisioning with its Unified Network Services. However, the cloud in box approach looks very similar to other vendors.
And Cisco UCS, in a physical box: http://forums.legitreviews.com/about34174.html
All the vendors talk about how their platform is revolutionary, and spew great volumes of excellent rhetoric to motivate the masses. However, with this approach what we are going to end up with are silos of clouds that are incompatible. We need to get back to basics and if these vendors really want to provide some revolutionary value, lets create automated provisioning software that can work across all hardware platforms. This means standards that to date have been very elusive to the entire industry. Lets create workload management solutions to manage “all” the software that might be deployed on the cloud. Then create network management solutions that work across all of my various vendor network assets. Accomplishing that means standards…that is where we need to focus otherwise instead of server sprawl, we will have cloud sprawl. Admittedly, cloud sprawl would place companies in a much better position than most are today with low utilization, manual provisioning, splintered monitoring, and the manual recovery procedures that are typically embraced. It will be a long way from what should be achieved and what will be needed to continue to meet a growing demand for always-on software services, compute platforms, and network connectivity.
So cloud wisely, as we condemned the mainframe model as closed and costly, we may do the same with these clouds in a box. In general, the industry has moved away from custom hardware appliances, as they are costly, requiring specialized training to use and support, with interoperability fairly low. As it stands now, companies will need specialized cloud support teams for each cloud system. That will be expensive, but weigh that against elastic nature of the cloud approach and I think it’s still a compelling proposition, even with potential of creating islands of clouds. Although, the strategy is more than get a cloud in a box and migrate everything to the cloud. It requires understanding multiple vendor private cloud interoperability, private to public cloud interoperability, total enterprise capacity planning, and then defining an approach that supports a unified management and monitoring view of it all. These offerings are a step in the right direction, but no where close to what I would consider a revolutionary destination.
Remember, luck is made, so make good luck as you cloud.