On Î¤ÎµÏ 05 ΣεÏÏ 2012 23:44:08 Chris Angelico wrote:
> On Wed, Sep 5, 2012 at 7:40 PM, Achilleas Mantzios
> <achill@smadev.internal.net> wrote:
> > (single master, 80+ slaves in 80+ vessels in the 7 seas (80+ = 80 and growning))
>
> Cool!! How do your nodes communicate with each other? Is it an
> off-line resynchronization, or do you maintain long-range (satellite?)
> comms?
Hello,
our topology is star-like. The system is based on good ol' UUCP
running on top of either ISDN lines or (as of late) over TCP/IP.
It is asynchronous and off-line by design. Vessels connect to the central master server
and get all their data, receive replication updates, and also send their data
to the office (central master cerver).
UUCP does the management of the queues (for the unitiated, think of UUCP as
something like JMS or AMQP or even better like JMS (API)+AMQP (wire protocol))
The comms (ISDN and TCPIP) are all done of course over a satellite service
(very expensive, so compression and minimal data replication were/are and will be
major concern)
In the case of ISDN, the billing is by time, so clearly this had to fit in the off-line category.
In the case of TCPIP, the billing is by data size, but we use that under UUCP
just like the ISDN off-line asynchronous mode.
Vessels can operate without connection to the office, and vice versa.
>
> The system I'm setting up at work kinda pales in comparison to that.
> It's designed to scale "to infinity and beyond" (and that quote is
> kinda appropriate, since we run this all on Debian Linux), but at the
> moment, all the testing I've done has been on a half-dozen
> off-the-shelf Dell laptops. But the same applies; we want absolute
> guaranteed reliability, so we NEED a good database. Postgres all the
> way! (Plus we need bindings for C++, Pike, and PHP, and I'm a lot
> happier with Postgres than several other options in that area.)
>
> ChrisA
>
>
>
-
Achilleas Mantzios
IT DEPT
