Geoff
12/29/2014 6:08:00 PM
On Mon, 29 Dec 2014 09:29:15 -0800 (PST), xeon Mailinglist
<xeonmailinglist@gmail.com> wrote:
>On Monday, December 29, 2014 5:08:21 PM UTC, Geoff wrote:
>> On Mon, 29 Dec 2014 08:18:57 -0800 (PST), xeon Mailinglist
>>
>> >I have this question, and I am trying to understand how I can calculate the availability:
>> >
>> >The company has two front-end web servers that provide some redundancy in order to ensure service availability. When a client accesses the service, any of the servers can be contacted (this is transparent for the client), and if one server is not available, the other will be used.
>> >
>> >a) If the 2 front-end web servers have a reliability of 0.9, what is the availability of the on-line store (measured in terms of the probability that the service will be up and running from a client perspective)? (Consider that all the other servers do not crash).
>> >
>> >My answer is 0.9^2*100. Is it correct?
>>
>> What is the numeric result? Is it logical that two servers with 90%
>> reliability each, when combined in the manner you describe in order to
>> increase reliability as viewed by the customer, has a total
>> reliability that is less than the individual reliabilities?
>>
>> What does Queuing Theory have to say about this?
>
>I was looking to this question, and I think this question has no answer because reliabitily is not the same thing as availability. It is not possible to convert reliability to availability.
>
>What do you think?
Yes it is possible but you can't use a reliability computation to
compute availability. The reliability of the system is a factor in
availability but it isn't the only term of the equation as you have
discovered. What you have computed is the series reliability, not the
parallel reliability.
The only time a customer will detect a failure is if both servers have
failed. What is the probability that a customer will find that
condition on contact?