Cloud Management - A Continuous Perspective

As I think about managing resources and services on the cloud (and yes I realize that one persons resource is another's service but...) there's a couple of thoughts I've been sharing with folks and Sun and our customers.

Clouds are a great example of applying "continuous architecture." Continuous architecture is a subject that has a basis in architecture but its really the cross section of biological thinking and the human task of creation. It's the notion of complex adaptive systems applied to something that historically might have been viewed as static (though buildings and houses are anything but.)

In the cloud, static really doesn't apply. A workload might be on one server one minute, on many the next, or on another in the next instant. This may or may not be apparent to the "developer" or service designer, or even the operator of the cloud, in fact its most likely not.

So how is this accomplished? For years we've (via dynamic infrastructure, JxSON, and other projects) thought that workloads are a composite of elements -- to run they require vLANSs, IPs, CPU, Storage, etc. As computer system technology continues to mature, so do the features, especially around the deployment, management, and virtualization technologies that enable much of the cloud today. Anyone that uses GridEngine or other grid/HPC applications -- this is old school. Yeah there's a resource manager, resources are mapped by the manager process, and they run, report back results.

But how does one assemble a bunch of virtualized resources that need to be brought together to meet the needs of a service? Are all the sub-elements assembled via a model and "put on the stack" or do you do this at runtime? How long does this process take? How does a service know when to return its composite resources back to the "pool?" Does it have a default time to live? The model map of the resources need to ensure that they are providing the level of service required by the service.

service_queues_01.jpeg

I was speaking to a customer the other day and every service by default is deployed in their DC with a 30 day TTL (time to live.) You can "purchase" an exception... I won't get into the downstream effects of this thinking -- and their are many but this forethought is significant.

But it illustrates these two processes....the consumption of composite resources by the deployment (run) requests for services.

Comments:

carolaja's blog entry dances around the issue of cloud efficiency. How often are resources 'returned' to pool for reassignment? At a given service demand level, a set number of resources are needed. As the 'resources in use' pool increases or decreases in size, the overall efficiency of the cloud model can be measured. Quickly restoring resources to the 'available' pool will increase the efficiency of the cloud model.

Cloud computing is the latest, cool, cutting edge, buzzword technology. Will it be the next definition of computing over the coming decades? Or will we look back and marvel at how nifty it was, but failed to drive a market adoption? Sun (and their competitors) need to design cloud computing from the business up, instead of from the technology down. Make it do more, better, faster, and cheaper. That will drive efficiency into the computing business - which will ultimately drive customers to the technology.

Posted by Martin McFly on April 15, 2009 at 12:13 PM PDT #

EM- hey hows it going? :)
So I agree with your points. I'm not specifically addressing adoption in my blog entry, though highlighting a shift in thinking that is occurring in the industry. The shift from deploying as a long lived activity to something that is done on demand, and continuously. Now old apps (or IT departments) might not ever move to this model, but many developers would like to just "attach" to an app container and interact with it as if its local, not-local --doesn't matter. Importantly the model has to be modified to include additional meta-data as the apps move from dev-mode to production.

Any how are you??

Posted by Jason on April 20, 2009 at 02:53 AM PDT #

Post a Comment:
Comments are closed for this entry.
About

Thoughts from Jason Carolan -- Distinguished Engineer @ Sun and Global Systems Engineering Director - http://twitter.com/jtcarolan - http://archmatters.wordpress.com

Search

Categories
Archives
« April 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
   
       
Today