Ken -
Lee asked about the same question on Dell getting into high end service, I replied in
exchange2000.com
My guess is that they won't go there if they can get the right results in a different way.
Is the support cost at startup or spread out?
The cost is spread out but is figured on a per-box basis for these calculations. Remember that costs for capitalized equipment are also 'spread out' by the depreciation schedule. Much of the actual service expense comes early in the cycle (planning and design).
Now the 'system availability' figure of 99.95-99.99 represents the time 'outside' of the maintenance allocation.
The numbers I was quoting include all downtime, scheduled or unscheduled. Tandem systems are the best in the business with downtime in the seconds per year from any cause. They routinely achieve 99.999 and have done 99.9999 for some very critical systems. VAX Clusters are guaranteed via service level agreements to 99.99 availability, and that includes scheduled or unscheduled.
S390 class systems do not all achieve these availability numbers, and as you point out, their availability numbers used to only consider times when the machine was 'supposed to be available'. They took so much heat for that in the mid-80's that they developed a number of configurations which provided online maintenance and upgrades to allow 24X7 operation.
A broad study of NT servers in large corporations (including MSFT's IT department which uses more than 1500 CPQ servers for internal systems) showed that as you say, the HW is pretty reliable but that the procedures to maintain and update the systems required frequent 'loss of service' events, on average about 2 events per month with about 4 hours loss of service in total.
If you don't upgrade the HW or SW on the server, don't change configuration, and don't make significant changes in security, then these servers can run reliably for a long time. Most large corporations don't have systems that meet those conditions.
Don't the big systems have the same problem?
Clearly not in the case of Tandem and VMS. I watched a demo in which every piece of HW and SW in a Tandem system was replaced while the system was executing a benchmark, and it never dropped below the required response time or performance level. VAX can do the same thing in a big cluster - the cluster needs to be big enough that loss of a single node does not affect performance by much.
Here again, IBM is not the availability leader but have produced some systems which can hit these targets. |
