Archive for the 'Cloud Networking' Category

Cloud computing and CDNs

Normally, cloud computing services and applications are delivered to users through an Internet connection. This is one of the pillars of cloud computing, making it distributed, accessible and affordable from wherever and, increasingly, via various wireless devices. But what about organizations that want something more than a “best effort” service delivery through the Internet?

Well, obviously there are several alternatives, including managed networks, e.g. MPLS, ATM and, in the future, many other types of virtual networks that run across  multiple physical networks or substrates.

But then there is the possibility of Content Delivery Network or CDNs like Akamai and Limelight Networks, which promise a better than best effort service or content delivery. Although the CDNs use in fact the Internet as the mechanism for carrying traffic, they strategically distribute replication servers (sometimes called Surrogates) in the network that replicated content stored on a origin server or servers farms.  Users accessing content from a particular server are directed to the most appropriate “surrogate” based on multiple criteria, including distance, network congestion, etc., determined by a load balancer in the network that calculates the most efficient delivery route. CDN providers use a combination of technologies to provide better than best effort service delivery, including the distributed surrogates, or caching servers, mention, but also by using different proprietary protocols/algorithms than the native Internet uses for inter-network communications, i.e. the Border Gateway Protocol, and by reducing drag caused by TCP multiple round trips to set up and tear down connections.

It is interesting to note that already some cloud providers have started to integrate CDNs into their products offering. This includes for example Rackspace that offers a storage solution called CloudFiles that is integrated with the Limelight CDN. Through the CDN, content can be distributed, cached and shared in edge locations throughout the world – so that users gain access to content from a nearby surrogate.

It will be interesting to follow this trend as see if and how more and more cloud service providers will integrate CDNs to their service offerings.

Networking technologies in Cloud Computing

Recently, I participated in a research study under the auspices of Eurescom, working with research colleagues from Telenor, PT Inovacao (Portugal Telecom – Innovation) and Orange Labs (France Telecom). Our objective was to analyse Cloud Computing as both a technology concept and service delivery model, especially from its networking perspective, and its implications for telcos in general. We studied the current and promising networking technologies used in the Cloud, internally and externally, including WAN technologies and Data Center interconnections.

Long-distance interconnections (WAN and MAN) between data centers are obviously based on IP standards (over ATM, Ethernet, SONET/SDH) and, more recently, on MPLS with QoS and native interconnection capabilities. For high bandwidth and ultra-low latency, DWDM (Dense Wavelength Division Multiplexing) appears to be very promising as a future high-performance WAN transport technology – mainly due to its capabilities of multiplexing multiple optical signals and being protocol and bit-rate independent (agnostic). Further on the horizon, new WAN networking technologies are still on the research stage, including the concept of Lambda networking that promises low-cost, high-capacity circuits in long-haul and metro systems.

Inside data centers , however, where network servers, storage systems and network nodes/elements are interconnected, three LAN networking technologies are prevailing:

Where, currently, Ethernet is the most frequently used. An important trend today is to deploy 10 Gb Ethernet (10GBE) equipment and networks – extending Ethernet’s capacity and support for more traffic patterns. FiberChannel is mostly used in scientific computing and storage area networks (SAN) whereas InifiBand is almost exclusively deployed in scientific and engineering simulation networks, e.g. using clustered servers.

Obviously, there are even more networking technologies and protocols available for supporting the delivery of Cloud services. This is an ever-emerging ecosystem of new innovations and improvements that will continue to evolve in multiple directions. For the Cloud vendor or service provider, it is, however, necessary to understand the strengths and weaknesses of individual networking technologies and how they most effectively can be applied in their current surroundings or technical infrastructure, whether that is inside a data center or in the transport sphere.

Network bottlenecks in Cloud Computing

Many enterprises are understandably reluctant in moving their core applications to the cloud, primarily due to security issues but also due, perhaps equally, to concerns of poor network performance. According to a recent report from the Yankee Group, many thought leaders, including Trend Micro, Cisco and CSL, say the issue of latency and poor performance is, at least temporarily, hindering the adoption of cloud computing.

This is not surprising. Quality of service delivery in the Cloud is intrinsically integrated with the network, its infrastructure and capacity. As migration to the Cloud continues, network operators face increasing challenges of upgrading the network infrastructure. This includes fixed infrastructure, including last-mile and first-mile as well as mobile networks like 3G and 4G. Many operators are already challenged with current unsatisfactory ROI – mainly due to flat pricing structures and “all-you-can-eat” data packages. How operators can justify increased investments in network infrastructure without changing pricing models remains difficult to see – unless they are, perhaps, also the Cloud provider themselves.

This is in fact the strategy Google seems to be pursuing by gradually increasing their network infrastructure possession. The latest example is their intention to connect up to 500.000 homes with 100Mbps fiber optic broadband connections (fiber-to-the-home), directly competing with the traditional telecom networks providers like Verizon. AT&T and Comcast. Google wants to provide rich Internet applications directly to the user – from the Cloud – eliminating network latency bottlenecks as much as possible. This probably includes bandwidth hungry high-def video applications, VoIP (Voice-Over-IP) and, of course, virtualized desktops a.l.a Google Chrome OS, where the desktop is actually being transferred to the Cloud and run from a lightweight network operating system (e.g. Chome OS). Google clearly foresees that all, or most, application will be run from the Cloud. Most likely, in my opinion, their vision will materialize in the coming years.