"Richard Burrows" <rich
...@rburrows.demon.co.uk> a écrit :
>I am looking for help in the specification of two servers, an Application
>Server Host and a Database Server, to run a product which is a (sort of)
>content management system focused in on educational use.
Content management systems don't get that sort of hit rate. If there's
a high-traffic site, and a CMS, then you might want to split the
servers (otherwise the CMS dies under the total load, and you lose the
"admin console" when you most need it)
For that hit rate, I'd be using more boxes anyway.
Dump MySQL !
> For budget
>forecasting purposes I am trying to spec a machine for a new version of the
>product which is not due for release until later this year and for which
>their are no published specifications.
8-)
>In the past the vendor has stated that it will utilise an "2" tier
>architecture base around industry standard components. I am assuming that
>the configuration may be something like:
Oh, you're using "foobar 2000" !
WTF is a '"2" tier architecture base around industry standard
components' ? That could mean absolutely anything.
>Application Server Host
First of all, I don't like Application Servers (philosophical rant
available). Get a good generic scripting platform and build out of
that, plugging packaged componentware in where useful.
IMHO, do the whole damned lot in Servlets and a bit of JSP for the
vague and wooly bits.
You don't mention a web server. If you're architecting a big system,
it's worth dedicating boxes as large and capacious servers of dumb
content non-interactive content, especially if you're offering a lot
of static content (image library etc.)
You might also want back-end components; either on the COM/DCOM/,NET
model, or as SOAP out to Servlets. Even if this is the same technology
as the front end servers talking to the punters, it's worth doing
complex back-end 3-tier stuff on a separate box.
>Windows NT 4 or Windows 2000 Server, or Red Hat Linux
>Java Development Kit
>Apache Web Server
Apache is happier under Linux (or BSD) than Windows. Oracle is
affordable if you can make that .ac.uk address work for you.
>Based on the current usage statistics I predict that the load on the system
>in three years time will be as follows:
>Maximum file hits/hour 500,000
>Maximum requests/hour 50,000
>Maximum visits/hour 20,000
>Maximum concurrent connections 500
>Bandwidth MB/hour 750MB
>Individual users 6,000
Hmmm..... That's not many users to generate that sort of traffic.
Either this is an on-line version of "Who wants to be a millionaire"
and the bursty traffic from hell, or something in those numbers
doesn't stack up as an average.
That sort of bandwidth suggests streaming video ? If so, you're
looking at big bills and possibly outsourcing its hosting to a shared
host in an enormous bunker somewhere. Even Akamai and edge serving is
cost-effective now.
Watch out for Moore's Law. 3 years is a _long_ time in hosting budgets
and traffic forecasts. You need to size around a shorter timescale and
prepare multiple forecasts with optimisitic and pessimistic ranges.
If you're managing this under PRINCE, you have my sympathy (I'm
working with Lottery funding right now). They just don't get the
notion that a rigid 3 year forecast is a bad idea on the web
Are your users "Friends of Janet" ? This might affect your
architecture and bandwidth needs. It tends to encourage an attitude
that everything (rack space, bandwidth) is free, which is unrealistic,
but handy for the initial stages of a project where the estimates are
vague. Out in the commercial world, server height is as important as
bandwidth, in terms of bottom-line costs. 1U server height can be a
big cost factor (take a look at Penguin Computing's Relion servers)
>I would welcome any advise on the specification of an enterprise system of
>this scale, based on comparable architectures using NT or Windows 2000, or
>Linux, or the address of resources where produce my own calculations.
You need to rent some geek-time from someone who builds these
regularly (I don't) There are all sorts of bizarre issues, like DNS
architecture, that start to be significant at this scale.
On the whole, I think we need more information on the function of the
site before commenting further, those volume figures look high
(especially for those few users) and there are probably some important
technology dependencies behind the scenes (i.e. is there anything that
_needs_ a copy of Word, or an XrML License Bank backstage ?)
--
Smert' Spamionam
