Amazon.com has quite a different challenge than my client. Let;s say they have 20 satelite offices. Each of those offices has 25 people using handhelds. Each office has a router with a dedicated redudant T1 to the main office. Hendhelds are all the same and a failed one can easily be swapped.

You are correct that they are subject to natural disasters and the like. And should one hit, they will have that problem. But they are not subject to failures from their own hardware or software.

Whatever they do has to have the same level of hardware and software reliability. I think they will leave the laws of nature out of it.

I am sure any of us here of a certain age remember when file servers like Novell were so rock solid they would run years not only without a problem, but without ever needing a reboot or the like.Microsoft has succeeded in lowering everyone's expectations.


>>I am interested in opinions on the following scenario, which I have to reply to and advise a client.
>>
>>First, a disclaimer: I know that absolutely everything can be done in absolutely every platform.
>
>That is not correct, but I'm pretty sure you already know that, and you're just exaggerating for effect.
>
>>If you wish to evangelize for a platform, this is not for you. I need to provide info based on the idea of offering solutions to problems that will best serve the client.
>>
>>The current system has handheld devices in the field. These devices run a terminal emulator and talk to a Unix server. There are disparate locations where groups of people with these handhelds do the equivalent of taking complex inventory by answering a series of questions on a text-based interface in the term emulator, and the results are stored in a database on the corporate Unix server.
>>
>>In addition, at each regional location (where the handheld users are) there are numerous users running reports and updating the data, either the actual data from the handhelds or related data.
>>
>>Then there are users at corporate mostly running reports.
>>
>>Then system has been running since 1999 with 100% uptime. All user processes are accessed 24x7 and it is mission critical that 24x7 access for reporting and especially data entry from handhelds be available. 99.999% is not good enough. They have government contracts and there simply can never be downtime -- ever, ever, ever, ever, no matter what. It might be surprising, but the data is in a variety of formats -- old FP unix tables.
>>
>>The system works great, but of couse allows for no modern interfaces. Adding reports or features is painstaking, and a web user interface so customers could enter or reviw some data themselves is not practical. And they are struck using forever an old version of Unix because of the FP tables.
>>
>>Now I have been asked to answer the following questions. Obviously I do not expect some of these to be answerable here, but to give you the picture:
>>
>>1. Should they move to a more modern platform to enable web interfaces, or should they keep their old system becauise of its rock solid reliabilty?
>>
>>2. If they should move to a modern platform:
>>
>> a. what should it be?
>> b.how can we ensure the same level of software reliability? Don't GUI's and wen and the like introduce an element of uncertainty that we avoid with our text system?
>> c. what would be the best data storage option?
>> d. what development platform would make the most sense for backoffice, in-office, and handheld-field use?
>> e. if we use anything windows based in such a system, are we introducing reliability risks?
>> f. how long and how expensive to do?
>>
>>3. if they should keep their current system because of its relability:
>> a. is there some way we can get web capabltities without changing the text-handheld system or the data storage?
>>
>>I have formed my opinions, but I wish to hear what some of the experienced consultants here have to say.
>
>I find the 100% uptime claim difficult to believe (since 1999, that's 12+ years). Five-nines reliability - which you say isn't good enough - is less than one second of downtime per day, about 5 minutes per year, to put things into perspective.
>
>There are many ways that a geographically distributed system such as you describe can fail. Some major disasters have happened since 1999, such as widespread power outages, hurricanes, 9/11 etc. If your client truly has implemented a geographically distributed system that has remained 100% available through these, and all the other mundane problems commonly encountered, they are arguably world leaders in high availability (HA). Even current cloud providers such as Amazon and MS Azure neither promise, nor in practice come close to, those kinds of service levels. I'm not certain even the Google search engine has been that reliable in recent years.
>
>If your client's primary criterion is 100% availability, then scrapping their existing proven system would be hugely risky, with no guarantee of getting an equally reliable replacement on a "modern" platform. I understand there is a so-called "middleware" industry that seeks to provide modern interfaces to legacy systems, but I bet they won't provide service level guarantees suitable for what you say your client needs.