Companies are now focusing more than ever on the need to improve energy efficiency. In addition to the cost of energy, a new challenge for them is the increasing social pressure to reduce their carbon footprint. Commercial electricity consumption is a major factor in rising atmospheric CO2 levels and data centers are a significant part of the problem. While energy costs are rising and data center equipment is stressing the power and cooling infrastructure, the main issue is not the current amount of data center emissions, but the fact that data center emissions are increasing faster than any other carbon emission. For this reason nowadays there is a big interest in “Green” data centers and supercomputer centers. In this scenario, the research community is being challenged to rethink data center strategies, and add energy efficiency to a list of critical operating parameters that already includes service ability, reliability and performance.
Consolidation and virtualization can be combined to reduce the management complexity of large data centers as well as to increase the energy efficiency of such a system. But even in a scenario where the resources are consolidated and virtualized, utilizing all the capacity of the components that are switched on (and consuming power) is not always simple. To determine a set of applications to be collocated in a node to perfectly fit and exploit all the resources of the system is a hard problem to solve, especially when tenths or even hundreds of nodes and applications can be found in a data center. Furthermore, the fact that the demand associated with each system resource for a given application may not be related in any way to its demand for other resources (i.e. an application with a large memory footprint may not be very demanding in terms of CPU power) which creates a structural problem requiring constraints to be relaxed in order to overcome it.

To solve this problem we have a poster in the ICAC conference held in Chicago last June.
In this paper we present how these two simple and wellknown techniques can be combined to dramatically increase the energy efficiency of a virtualized and consolidated data center. Increased energy efficiency is obtained through the introduction of a new approach to the consolidation strategy by combining: memory compression and request discrimination. Combining these techniques enables an important reduction in the amount of active nodes required to process a web workload by dynamically classifying and shaping the workload, without degrading the offered service level. Furthermore, when the system eventually gets overloaded and no energy can be saved without loosing performance, we show how request discrimination can still improve the overall value obtained from the workload. The two techniques were separately studied and validated in a previous work to be now combined in a joint effort. Memory compression is used to convert CPU power into extra memory capacity to overcome system underutilization scenarios caused by memory constraints. Request discrimination is used to characterize web clients by predicting their class and the value they will have on the system. Our experiments performed on a real workload, obtained from a top national travel service exemplify the dramatic improvement these thechniques offer in energy and performance efficiency.
The main contribution of this article is to demonstrate that the consolidation of dynamic workloads does not end with virtualization, but there is even more to consolidate when energy-efficiency is pursued. We will present two alternatives to rescue resources that consolidation does not currently capitalize on while using virtualization.

link to the poster - link to the extended version of the paper.

The Gartner Group recently presented their vision of the top ten disruptive technologies  they believe will change the world over the next four years:

•    Multicore and hybrid processors
•    Virtualisation and fabric computing
•    Social networks and social software
•    Cloud computing and cloud/Web platforms
•    Web mashups
•    User Interface
•    Ubiquitous computing
•    Contextual computing
•    Augmented reality
•    Semantics
Some of these are of particular interest to our EEDC course in the CANS master course as can be reflected in our EEDC’08 workshop today.
Today is the last class and let me again remark that in my opinion “Cloud  Computing” (remember that it means different things for different readers, however it appears everywhere) will grow considerably over the next several years as it enables the sharing of storage and computing infrastructure with others via the Web, including the maintenance and operations (cost-effective approach). I also think we will see application-specific instances of cloud computing on the Web grow considerably, like for example Salesforce has done. Taking in account the list above, and following our class discussions, some of my thoughts are that  multi-core processors and virtualization will mean that companies will need fewer servers, space and cooling systems (consider Oriol’s presentation). Also, applications  can be more easily moved from server to server, or moved from data center to data center and even to an outsourced data center. It could mean that the need for a centralized IT department in a company will go away completely. Companies will no longer need to own/maintain the data centers that they use to run their applications. And with no need to touch a data center directly, will there be no need to have their own IT staff? Of course the IT industry and research will still need IT experts like you.

As you know during this EEDC course I tried to understand how research is going to change IT and transmit this to you. Let me quote a phrase from a post by  J. Anderson that reflects my vision: “The key to your future success is to understand how IT is going to change and what you need to do to change with it”.

I hope that this course has been useful to all of you.  I want to wish you lots of luck in your future careers.

Currently Barcelona is hosting the Open Grid Forum (OFG23). OGF23 is one of the largest and most important events taking place related to Grid and distributed computing technologies. One of the keynote presentations was given by Dr. Werner Vogels, the Vice President & Chief Technology Officer at Amazon.com, who is responsible for driving the company’s vision for technology, and must continuously enhance the innovation for Amazon’s customers on a global scale.
I was very interested in attending this presentation (even though it was at 2:00pm, Barcelona lunch time ;-)) because we talked a lot about elastic computing in our course and I think Werner Vogels is a very important person in this field.
The presentation was very interesting (even though a little bit commercial in my opinion). The title was “AHEAD IN THE CLOUD. The power of Infrastructure as a Service” and explained some very interesting things that we already discussed in the course.
One of them, was related with the exercise that we did in the course to try and start a new company using the Amazon Elastic Compute Cloud. Remember that in general we only consider the cost of the time to install the hardware (in terms of Euro and time).
He confirmed what we discussed extensively in the course; that today, with the arrival of cloud computing, it is no longer necessary to make a big investment before a company initiates their business.
However, in the course we did not discuss the steps that a company is required to do before a new service can be turned on. He tried to convince us of the benefits of using elastic computing by looking at the traditional chain of actions that have to be done by a company to get some new software running (develop, test, release, install, configure, operate,… which he calls “Software as Bits”), when compared with the “Software as services” model that cloud computing allows. This is a really big issue to take into account also.
I hope that the presentation will soon be available on the OGF23 site and I will notify you to take a glance at it.