Thursday, 16 April 2009

Factors for consideration in the performance design of Cloud Infrastructures

The evidence of performance issues in today’s cloud are seen in the public domain through outages of service , loss of business through unavailable 3rd party online commerce payment services and even closure of site services with impact on the subscription customers data and business held in these sites. More mundane but nevertheless an issue is the slow performance of high usage on popular web sites and the “drag” factor on user latency time at the keyboard and mobile.  All these issues combine to create a less than satisfactory Quality Of Service QOS experience for users and providers alike. 

The introduction of cloud computing concepts introduces abstraction at the edge network and the technology tiers of infrastructure, applications and business process provisioning.  Often the argument made against federated service orientation is the increase in complexity of distributed IT services when the goal is in fact to reduce complexity and simplify the user experience of the service at the edge boundary.   Data, storage, servers, network and devices are all virtualized and aligned as a delivery framework that makes the selection, delivery and deployment of the user experience much more “plug and go”.     But there are factors that need to be considered in the compression of the time and cost of delivery set against the methods of abstracted and compartmentalizing a secure infrastructure to deliver the service.   

Amdahls' Law Rising

The experience of Amdahl’s law shows that the theoretical limits of compound use of more and more processes has a diminishing return on the performance enhancement of CPU processing speed.  But this effect is not restricted to processor design.  Other factors of overlaying additional abstraction tiers on top of physical and logical representation increases the number of addition computation and translation points of a system.  This is a complex issue of node count and stages of processing versus the use of appliance hardware and energy and space saving design which is at the forefront of new IT architecture and language design .  Much of the current research into standards for open VM containerization, federated presence location Identifiers and distributed storage standards such as DMTF OVF, IETF LISP, UDT / UPD and XMPP and of course the ubiquitous “standard Cloud API Adapter” are seeking to make this transition to a universal connection and service provision model.  

There are factors that need to be considered in performance design of the individual on demand services and the orchestration between services.  

  • Physical level
    • “Bare metal” and Virtualization of data base, storage cluster structures and servers
  • OS Level
    • The machine image and application containerization  and portability methods for replication, separation and transportability between distributed host environments
  • Message level and module level
    • The message standards granularity and API protocol selection for stateless and stateless communications
  • Process level
    • The virtual network overlay tunnel impact on performance ; the federated
  • Appliance level
    • The scale density and performance characteristics dependent on user systems awareness of the appliance
  • Device Level
    • The method of abstraction and degree of tight coupling to the host device and services 
  • Market Level
    • The commercial and contractual agility and effectiveness of the service and position in the industrial market supply chain


No comments:

Post a Comment

Note: only a member of this blog may post a comment.