Cloud Interoperability session

Benoit Tremblay (Ericsson) introduced the purpose and scope of the Cloud Interoperability session; that is to advance our understanding on the interoperability challenges and candidate approaches followed by several projects. He noted that many application types (such as gaming) are offered on the Internet through many cloud providers (like amazon, rackspace, etc.), but interoperability is hampered due to the multitude interfaces used by each operator. Candidate approaches for moving from isolated to interoperable cloud services were suggested to be standardized interfaces and Cloud federation; where the difference lies in the number of providers an end-user interacts with. However, common interfaces are necessary in both cases.

Azimeh Sefidcon, from SAIL FP7 project, started by describing their approach to Cloud Networking, named CloNe. In their architecture, both peering and hierarchical (related to federation) delegation of requests is possible. Then, she mentioned a set of challenges. In particular, ISPs don’t reveal network topology information. This makes difficult the optimal placement of Virtual Machines in different domains. Another challenge is the Information hiding from network operators to application providers.

The “cloud curtain” issue (no visibility) was also raised by Sergi Figuerola from GEYSERS FP7 project, suggesting that well-defined architecture & business models are needed in order to have an impact on market and even force “hyper-giant” cloud providers to make use of open technologies. But the main problem in achieving cloud interoperability, as was identified during the panel discussion, is that large operators (Google, Amazon) don’t care about interoperability. This is also evident from the fact that integrators (brokers) are the driving force for standardization. However, interoperability is known to help in achieving a competitive landscape in an efficient manner.

Another dimension of problems pointed out during the panel discussion is related to interoperation with network operators. Network QoS guarantees are not prevalent today and even when an ISP offers such SLAs, monitoring is necessary in order to identify the cause when something goes wrong.

Cloud Management session

Alex Galis, representing AUTOI + RESERVOIR research projects, highlighted the importance of management software in three types of shared infrastructures (network clouds, computing clouds and integrated clouds, or software-defined networks) and mentioned that similar challenges exist in all three cases (e.g. QoS management and autonomic operations).

Similarly, Laurent Ciavaglia from UNIVERSELF research project posed the question whether there are new, additional, or significantly changed requirements specific to cloud computing and/or cloud networking. It seems that there are many similarities when dealing with the management or other performance-related objectives (such as stability, scalability) of cloud and networking resources.

However, the complexity is greater when these resources are combined. For example, Ciully from GEYSERS mentioned the need for cross-layer optimizations, such as when computing an end-to-end path subject to IT constraints, which in turn creates the need for a Cloud-to-network interface.

Marcelo Yannuzzi, representing ONE research project, argued that Cloud computing will have dramatic effects on networking (e.g. Google is said to be working on a carrier-grade router for 6000 euros). Furthermore, he suggested that better coordination is needed between network & IT management systems, instead of integration.

One example of such type of coordination was described by Panagiotis Demestichas from OneFIT, where an ISP can form an opportunistic network across multiple Access Points in order to meet the increased bandwidth demand of the users/ applications in a particular area.