--- In firebird-support@yahoogroups.com, Daniel Rail <daniel@a...>
wrote:

> Hi,
>
> At February 7, 2004, 06:02, Werner F. Bruhin wrote:
> > Wouldn't want to contradict and/or argue with Helen, BUT if I

were you

> > I would make sure to setup a test environment with at least 30

clients

> > and go crazy on it, while on average you might only have 15

stations,

> > when it counts you will have 33 stations working - AND at this

point

> > the system has to fly!
>
> I have to agree. What needs to be done is proper load testing.
>
> > The test should help you and your company to determine what kind

of

> > server configuration you need to have and to ensure that anything

you

> > can tune on Firebird is correctly tuned for the heavy load of 30

plus

> > clients.
>
> I would compare between using FB SS and FB CS, either on Windows
> and/or Linux. The system would have a fast processor(or processors

if

> using FB CS on Windows) and enough RAM(1GB at least). Also consider
> using switched network hubs, instead of non-switched hubs, this will
> help in the network traffic.
>
> I have a customer with 22 concurrent clients with FB SS 1.5 on a
> Windows P4 3Ghz 1GB RAM. And, he is very satisfied with the
> performance, because the response is practically immediate.
>
> Also on a LAN, a sub second response by the application for a POS is
> usually enough. On a slower network connection(10Mbps and less), the
> response by the application shouldn't be more than 2-3 seconds. And

do

> realize that I'm referring to the application and not solely on the

FB

> connection time, since your application might be doing more than one
> DML statement to perform a certain task.

On a POS (for a Restaurant) any/all transactions should be sub second,
regardless of network - we had this on networks way slower then LAN.

By transaction I mean an order of e.g. a steak, a coke, etc, only
exception is probably the payment transaction if a credit card or a
room charge is involved as the authorization will take longer there.

Obviously all the timing should be measured on the client station,
even if it is done by human beings (it is easy to see if you have sub
second or seconds response time). If you have a tool to emulate the
user on a client even better, especially if you find that the response
time is NOT sub second and you need to change things and check it
again and again.

Would be interesting to see your findings.

See you
Werner

>
> --
> Best regards,
> Daniel Rail
> Senior System Engineer
> ACCRA Group Inc. (www.accra.ca)
> ACCRA Med Software Inc. (www.filopto.com)