As so often with new trends in I.T. prospective customers soon find themselves being besieged by salesmen, marketeers and industry pundits demanding to know what their plans are for implementing the technology. In an industry which relies on novelty and constant innovation for the continuance of its revenue stream this is probably unavoidable. But I, for one, am rather tired of a certain tone of voice which implies that if I have not chosen to jump aboard this particular bandwagon then I must in some way be failing in my professional duties.
In recent times I should have been investigating convergence, wireless, web services and outsourcing. Before that I should have been investing in ASPs: I wonder what happened to them? The truth is, of course, that these are splendid innovations and under the right conditions can save you money and improve efficiency. It is, however, necessary to look very closely at your particular circumstances before taking the plunge.
Consolidation – of storage and of servers – is certainly one of the issues of the moment. The claim made for storage consolidation is that it reduces hardware and staffing costs. This is based on the premise that large amounts of disk space are often left unused with directly attached storage and that there is a high administrative overhead in backing up dispersed data stores.
Server consolidation is thought to be the next logical step after storage consolidation. Again it is claimed that money can be saved by avoiding the purchase of many small servers. Now that network bandwidth is inexpensive locating all services in one central location can simplify administration.
So why has Eton not consolidated its servers or storage? Why in fact do we purchase ever more servers?
Servers fit for task purposes
Our strategy is to tailor server provision as close as possible to the requirements of each service we offer. The academic environment demands great flexibility, with new services being added and existing services modified, expanded or removed on demand. We need to be able to cost each service accurately and separately, and to manage them independently where required.
We purchase appropriately-sized servers for each task. Name services, domain controllers, directory catalogues, web services, email servers, user home directory services, and a multiplicity of administration and database services all run on their own dedicated hardware. These machines are sized and optimised for the service they support. This means that we can use very small and simple machines for many jobs and only need massive redundant disk arrays and multiple processors for the more complex tasks. We are currently installing new servers to support our Virtual Learning Environment, and can optimise the machines carefully to match the particular storage and performance requirements of the web application involved.
Our primary operational concerns are resilience and security. Most of our servers are duplicated for redundancy and services such as domain controllers can be dispersed as widely as possible. We can suffer a number of simultaneous failures without our users experiencing loss of service. Our network is based on a high speed ring with backbone switches serving the different zones throughout the School. We house servers in the same location as the backbone switch closest to their major user departments. Servers are connected directly to the backbone for good performance, but are immune to network or power failures in other parts of the School.
Running many independent servers also has benefits when it is time to replace or upgrade machines. Our current array of servers was purchased over a five year period, and are thus due for replacement at different times. As we can consider each part of the provision in isolation it is simple to calculate the cost of any upgrade and the effects of any change of requirements. It is also possible to redeploy components to other services as required and to carry out a continual rationalisation without major implications to the service as a whole.
We see no advantage in consolidating our storage even with such a plethora of servers. We prefer to optimise directly attached storage for the service concerned; fast disk arrays where appropriate – slower smaller individual disks otherwise. Our backup strategy is only to archive regularly stores that change routinely. For many servers a complete rebuild is more efficient.
Though a school such as Eton might not be under quite the same financial pressures as commercial organisations our belief is that the efficiency and the effectiveness of the service should override cost considerations in most cases. Interestingly the advocates of consolidation sell it not only on cost grounds but also in terms of efficiency gains. Such gains are often difficult to quantify or attribute, and should in any case be studied carefully.
Are we totally opposed to consolidation? Well – no. In administrative terms it certainly makes sense. We standardised very quickly on one platform, with all servers administered centrally. We also consolidated our backup strategy, archiving our dispersed servers to large networked backup devices. This strategy has enabled us to manage a large number of disparate servers with a relatively small team of highly qualified administrators.
I don’t know if a commercial organisation can learn anything from our experiences, but I would certainly recommend that anyone considering consolidation makes a careful study before rushing out to buy.