Real time service in the Cloud

The IRMOS project is taking Service-Oriented Infrastructure to the next level with an innovative approach that provides quality of service guarantees, as required by real time interactive applications, when executed on a service-oriented environment. We talk to Dimosthenis Kyriazis and Stuart Smithson to find out more...

While it is true to say that Service-Oriented Architecture has been widely adopted as a concept in servicing the IT industry, the adoption of Service-Oriented Infrastructure (SOIs) has lagged some way behind. That has changed somewhat with the evolution of more robust grid and cloud infrastructures and the acceptance of virtualised servers and storage, but one issue remains that prevents adoption on an even larger scale – the lack of real-time (RT) capabilities in SOIs.

In essence, it is generally accepted that SOIs are not dynamically configurable and adaptable to RT requirements, while timing and interaction issues are not formally expressed in them either, making any dynamic allocation of infrastructure resources harder to implement in terms of guaranteed Cloud-oriented services on demand.

It is a problem the IRMOS project (Interactive Realtime Multimedia Applications on Service Oriented Infrastructure) first began to consider early on in 2002, when researchers had identified a gap in the market left by this lack of RT capability on grid infrastructures. By this time, of course, the Cloud paradigm had emerged and so the IRMOS proposal was produced, which set out to develop a service oriented Cloud computing infrastructure. The success of the project depends inherently on the Cloud and it must build a Cloud infrastructure consisting of resources which are utilised, from network links to storage units. This is in essence the way in which the notion of Cloud computing actually came together. In this context IRMOS will bring in RT functionality to enhance end user experience and to deliver services that are both cost effective and quality-of-service guaranteed.

Heading up the technological development within IRMOS is Dimosthenis Kyriazis from the National Technical University of Athens and he is able to outline exactly what all this means in terms of development, as well as the greatest challenge he faces in making it happen. “I sum up what IRMOS is working towards as service interaction,” he explains. “This means that we allow services to interact in a very dynamic way, to interact continuously and to span different domains, from theapplication level to the network, to the management level and into the environment. This is why we call it a “Service-oriented”, Cloud computing environment.

“But the most challenging concept in achieving this is what we call the ‘new generation’ of applications. This is about interactivity; even games are interactive now – so any platform, any environment must be too. So Cloud computing service infrastructure should serve platforms facilitating interactivity – it is the applications that are interactive and the infrastructure that services them should facilitate that interaction. “So our vision is to facilitate RT interactivity, but not this alone – and this is what will make IRMOS such a success,” continues Kyriazis. “We also aim to provide qualitative service guarantees.”

These guarantees are at the heart of the project and here Kyriazis uses Google as an example of what he means by this and why this development is so important in terms of service delivery in the Cloud. “Imagine,” he explains, “that you are at home and you turn on your computer and you get a ‘page not found’ notice.

“You may check your internet connection but you would never imagine that Google was down or unavailable; this is called Quality of Service; it is about availability and reliability and this is fundamentally why Google is trusted. “Performance and results produce a high level of quality and quality such as this is very important. This is why IRMOS is working on this new generation of Service oriented Cloud computing environment as it provides quality and it will provide secure guarantees for this quality.”

By developing an infrastructure with guaranteed level of service, IRMOS is looking to change Service-oriented Infrastructure for good – and not before time. Before IRMOS, Service-oriented Infrastructures were not performance driven and this was a major problem when attempting to adopt service infrastructure for business. Quality of Service guarantees for infrastructures are essential in order to make it a business proposition and it is the provision of quality of service guarantees that makes the project most unique.

The question now, however, is how realtime functionality relates to Quality of Service and why this would be a focus for IRMOS? Firstly it is important to make the distinction between real time and ‘soft’ real time applications – ‘real time’ does not mean ‘instant’; it means time according to specific quality of service requirements. This ‘soft’ real time, according to IRMOS, acknowledges that there could be some problems in the running of an application that are acceptable and these are identified in the formulation of the service agreement and a quality guarantee up front. “What these applications actually require,” picks up Kyriazis, “are time constrained operational services. You need services to operate in a timely and constrained manner based on certain tasks that might have been set by an application site. For example, time-critical operation involves communication between services and, of course, the end user.

“It was this communication aspect that had a direct impact on the design of them virtual systems we have developed and these interactive systems did not have this before. This is what RT applications actually require and this is what IRMOS provides,” he concludes. In creating an interactive RT multimedia application and moreover, an application that has now been adopted, the real challenge facing IRMOS has been developing the infrastructure on which it operates – something that has involved research at the highest level. Kyriazis breaks these research challenges into six main areas:

1. Modelling of the application

Research was needed into the techniques for modelling the applications and for predicting and estimating application-specific infrastructure needs. “If we don’t know what the application needs, we cannot provide the infrastructure requirements or the Quality of Service levels required by the application,” says Kyriazis. All this modelling, prediction and estimation of the research needs are expressed with models.

2. Commitment of the applications

The biggest process for establishing service level agreements (SLAs) covers the relationship between the different providers in the Cloud – platform providers, infrastructure providers and service providers. “We needed to develop ways and means to facilitate the negotiation of contracts between the different entities; not only this, but we need this to be done in an automated way,” says Kyriazis.

3. Resource management

This process involved working on the management of resources that need to be assigned to certain applications, while estimating the isolation that was required. In essence this entails a multi-tenancy environment with various interacting applications within the same environment warranting a need for isolation.

4. Adaptation

This relates to the management and adaption of the system to failure. It is necessary to

establish recovery mechanisms in order to guarantee quality of service. “In the case of exceptions or violations, we have to detect and react in time,” says Kyriazis. This adaptation exercise is in relation to RT execution and prevents that execution from being interrupted.

5. Infrastructure

One of the most important challenges facing the project has been the development of the infrastructure. “You have to have an infrastructure consisting of computing, storage and networking that is real time aware,” says Kyriazis. “If so, after completion you can have

performance guarantees in terms of completion times.” These guarantees take into account different parameters – for delay and for latency. The challenges arise in terms of hardware and in terms of having to virtualise the resources.

6. Integration

There is no defined methodology, approach or precedent on how to have a complete RT solution/environment and because of this the integration of RT applications on to all levels of this environment, from the applications down to the infrastructure became one of the project’s biggest challenges.

“We had to appraise all the practical issues of performance guarantees, time execution, real time management of resources and so on and ensure that this was all integrated for the final solution,” says Kyriazis. In meeting these challenges, IRMOS has been working in an area where there are no established standards so it has been necessary to work on standards development.

Several members of the project consortium are currently on European standards bodies, while several new projects have identified specific IRMOS components, methodologies and services that they will now adopt in their work. It is indicative of the value IRMOS brings to the ICT environment and the impact the project already has on the SOI environment, particularly amongst those who have expressed concerns about the Cloud computing domain. This is perhaps best illustrated when the end-to-end nature of the project is considered. It provides developers and users everything they need to set up their application so that it runs correctly and successfully within a service environment of this type. It also provides the benchmarking of that, so that they can guarantee an SLA which is relevant to that application within the environment they need. It also enables them to negotiate that SLA and negotiate its technical elements down to the infrastructure so it becomes an end-to-end demonstration of RT applications in these environments.

Kyriazis believes this is a unique approach to SOIs and describes the approach IRMOS is taking as two-phase. “We have an off-line phase, during which you have a design time and within this phase you are able to develop an application for the IRMOS environment,” he explains. “Then there is the main phase, during which the applications are applied and executed, the resources are negotiated automatically and a reservation for the resources is then made. The whole process is, of course, monitored.”

Stuart Smithson of Xyratex, based in Portsmouth, UK, is the project coordinator for IRMOS and believes that the true innovation in the project lies in the fact that it moves all aspects of the infrastructure and the architecture to become RT oriented.

“Of course, this doesn’t necessarily mean hard real time, where response is in microseconds,” explains Smithson. “But the real time aspect is something which is always appropriate to the particular applications. We can accept lost frames here and there, small delays, but that is all part of the service agreement which is agreed up front, with what will be a paying customer in an eventual product that was generated by IRMOS.

“So the innovation, initially, is the inherent real time aspect of it. That links into the quality of service guarantees which are given, which drives the innovation in how we limit the use of resources by multiple customers effectively on a single resource,” he continues.

“By looking at what we generate from there, as individual organisations, in terms of intellectual property, we would then want to use and have some structure from other applications. From a Xyratex perspective, we have developed intellectual property, which allows us to control the usage of the storage media. “Traditionally, people would look at storage and simply throw in volumes of capacity to make it as adequate as possible, but what we are now seeing is what people are actually signing up for what they actually require and a guarantee that they will get value, while not stealing anybody else’s bandwidth or storage capability. “The same applies to the network and some of the processing elements. Each of them has got its own technology, created to both negotiate the level of service and guarantee the quality of that element of the infrastructure and that is then brought together as a whole and delivered through the SLAs to the applications.”

“Effectively, we’re saying: ‘How much of that resource do you want to use?’. Users can then plan their usage and that usage profile will be priced appropriately, be it at the weekend, during the night or during the day within the cloud environment. Eventually, any product that comes out of this in terms of a commercial deliverable would offer the best possible value for money for the resources required.”

Ultimately, IRMOS is providing the best utilisation for that infrastructure and dynamically allocating the resource requirements according to SLA demand in a guaranteed way, but minimising the level of over-provisioning. It is possible to run a real time application in many of these environments today, if you take over the whole environment and use it exclusively for your single application, but that is not a service.

Users need to be able to exist with other coherent applications and operations going on at the same time in a given infrastructure and, in those circumstances, they want to have an optimal/practical level of provisioning of the resources that guarantees what they need, but no more. That is something that was not possible in the past, in many environments (although it was better understood in the networking world than it was in the storage world) and it is something IRMOS now provides that no one else does.

This, of course, puts the project in a very strong position and it is already generating a great deal of interest among users looking to guarantee the level of service they require at a cost that reflects only what they need and ultimately use. “I don’t think that you would use the infrastructure if you didn’t have the guarantees,” adds Kyriazis. “This infrastructure doesn’t target specific applications; it is an infrastructure that is implemented to be used by anyone who wants a distributed infrastructure offering quality of service guarantees.”

Here Kyriazis uses a case study to illustrate how this usage may work in the real world. “Imagine that you are running a film production company,” he says. “The directors are shooting during the day and the colour correction in production needs to be done at night as the team need to see the result the next day to decide whether further changes are needed. This normally takes time but with IRMOS it is possible to do this task in two or three seconds with any change performed by the colourist seen in real time when decisions can be made there and then.”

Already IRMOS has three applications developed, demonstrating three main uses for the technology to be used in eLearning, visualisation and the film and automotive industries. All the future applications will have RT specificity. There will be no future interactions that won’t have this interactivity making it possible to have distributed applications, instead of having expensive hardware and this is the innovation.

For further information, please visit the IRMOS website here.

Published: Tuesday, 4th January 2011 by William Davis