Three Architectural Trends Effecting the Cloud
By jasoncatsun on Oct 15, 2008
There are three trends that influence the power of the cloud (and probably more!) The power of general purpose systems going beyond good enough to best in class. The increasing convergence of key architectural layers into software, and its resulting significance, including the operating system The network is no longer a passive part of this system, one can't exist without the other but they are worth looking at individually Deployment Architecture as “Software” Finally, the maturity of “Good Enough” Computing. Examples of this trend are the Thumper out performing storage appliances and Zeus ZXTM + CMT out performing a hardware based content switch. This maturation goes beyond “good enough” to highly optimal. For most companies, good enough at the right cost level meets their needs. This is a shinning example of how cloud infrastructure can help. So what is good enough? Good enough is seeing great to excellent performance out of an application on a general purpose server or system. A side-effect of the general purpose “good enough” strategy is that the inherent virtualization brings additional flexibility to bear that can contribute to overall efficiency in a cloud environment. In the diagram below, a compute node changes functions by overlaying a different persona on the same underlying device. In this “good enough” world a server might be a web server (just a boring node) one day and be a content switch and cache (perhaps the center of the cloud?) tomorrow (or a few minutes later!) There are several things that make this possible and functional: A standardized, meshed networking topology that is “cabled once” and adjusted dynamically this may utilize VLAN tagging, port-based VLAN control via a network provisioning product, etc. A systems management platform that is able to adjust the servers personality – transforms it from a web server to a content switch and applies the appropriate policy-based configuration A policy-based system that orchestrates other system changes necessary (else monitoring systems, routers, etc may be out of sync) from an deployment descriptor or instruction set. General Purpose to Multi-Purpose: Weighing Features and Complexity A key enabler to this approach is the power of general purpose. This has led to a convergence of features into a single key building block, which in turn is allowing an architectural convergence. OS-level virtualization, features in Glassfish that offer Java as well as other language support, and meshed network backplanes supporting both Ethernet and Infiniband are great examples of this. One layer of the architecture can now do many things. The operating system is another example influenced by virtualization. Before the OS was tightly coupled to the hardware and the applications it provided. Now there are multiple layers of operating systems with a layer of virtualization that helps map application requirements to a lightweight OS to another lightweight OS that manages the hardware. One may change a lot, the other may only change when the hardware lifecycle is about to expire. Further, a typical three tier topology can now be represented in a single physical server. This happens every day on the laptops of system engineers and now it happens in the data center. This also increases the level of complexity embedded into that building block. Security, reliability, maintainability, etc are all concerns when what was separate and distinct layers are now logically represented – if represented at all! This highlights the need for models, patterns, and well understand use cases in an increasingly compressed and dynamic world. The Network Is the Dynamic Computer Finally, the network is critical to this model. It is the abstraction and critical connection point between the resources necessary to deliver the cloud, from the end point (service access layer) and within the architectural layers. It also provides a critical layer of security defense. While the services may share common physical devices, the network provides the last mile of connectivity, and can provide features such as encryption and authentication at wire speeds and scale. Organizations will grapple with the coordination necessary for network changes to correspond with system changes. Today we see VMWare take control of parts of the network hence more infrastructure silos and differentiation being added to the overall data center structure. This increases complexity by not have visibility within these silos and may introduce more problems system wide (how many folks have had multiple DHCP servers clobber production services??) Networks are now cabled once and use policy-based systems to automate common tasks and provide guard rails for these changes. See Intelliden and Alterpoint for examples, or open Netconf initiative. See also project Stratus where the network is further embedded into the compute platform itself and not separate devices.
Technorati Tags: cloud computing