Database design and pl/sql vs external procedures

Question

hi,
My project involves predicting arrival time of a bus at a bus-stop, given statistical data of traffic patterns on the previous ‘n’(say 3) days, as well as the current location of the bus(latitude-longitude).
Given current bus location, I derive my distance-until-destination bus-stop, which must be translated into time until arrival.
Ive enlisted the triggers and procedures involved in making the prediction. Thse procedures especially the determination of perpendicular distances involve some complex trigonometric operations. I would like to know if my approach is correct and my database design is suited for the operations to b performed.
Will it be more efficient to implement the procedures as external procedures or as pl/sql blocks
This is my database design:
LINKS ( a link is the road segment between adjacent bus-stops)
LINK_ID                NUMBER      [PRIMARY-KEY]
START_LATITUDE          NUMBER     
START_LONGITUDE     NUMBER     
START_STOP_ID          NUMBER
END_LATITUDE          NUMBER
END_LONGITUDE          NUMBER
END_STOP_ID          NUMBER
LINK_LENGTH          NUMBER
BUS_ROUTE
ROUTE_ID               NUMBER
LINKS_ENROUTE          VARRAY(30) OF NUMBER
STOPS_ENROOUTE          VARRAY(30) OF NUMBER
TRACK(keeps track of current location of bus)
BUS_ID           NUMBER          [PRIMARY-KEY]
ROUTE          VARCHAR2(20)
LATITUDE          NUMBER
LONGITUDE          NUMBER
TS               TIMESTAMP
LINK_ID          NUMBER
START_STOP     NUMBER
END_STOP          NUMBER
ARRIVAL_TIMES(actual arrival times of the bus, updated by track)
BUS_ID           NUMBER     [PRIMARY-KEY]
BUS_ROUTE          VARCHAR2(20)
ARRIVAL          TIMESTAMP
STOP_ID          NUMBER
ETA (expected time of arrival)
BUS_ID          NUMBER
BUS_ROUTE          VARCHAR2(20)
BUS_STOP_ID     NUMBER
ARR_TIME          VARRAY(5) OF TIMESTAMP
Triggers and procedures
1)TRACK_TRIGGER
On insert/update of track determine which link the us is currently on.
Invoke a procedure that calculates perpendicular distance from current location to all links en-route (cursor on LINKS).
Results are stored in a temporary table. Select the link-id of the tuple whose perpendicular distance is MINIMUM. This is the link the bus is currently on. Place link-id, start_stop_id and end_stop_id in corresponding row of TRACK.
2)ARRIVAL_TRIGGER
update ARRIVAL_TIMES.and store in ARRIVAL_TIMES that start-stop id with the time-stamp of the current track record.
b)update ETA: Find the BUS-STOPS that come before the START_STOP of current track record. All these rows are deleted from the ETA tables, as the bus has already crossed these stops.
3)Prediction Algorithm Procedure.
Determine distance until destination for each STOP, 20 stops down from stop current location.
Determine current avg. speed of the bus over a 2 hour window, by dividing total distance traveled by time taken.
Calculate time-until arrival T1=current avg. speed * distance until destination
From the records of previous ‘n’ days (say n=3) find those buses on the same route that were near the link the bus is currently on. Again determine avg speed over 2 hour window and calculate avg. speed.
Calculate travel time T(i) = speed*distance until destination.     i.=2,3, 4…
The final predicted arrival time is a weighted sum of all T(i).
I hope Im not asking for too much, but the help would be greatly appreciated.
Thankyou,
Amina

Answer 1

hello,
actually i can manage ETA without a varray, since there will b a maxximum of 3 -4 values of expected arrival times at each stop. this can b done with separate columns.
though i dont quite understand how lag() will help me...from what i understand lag() is to access values of previous rows. but in ETA table each element in the varray(if there is one) is going to be the expected arrival time of buses on a particular route at that particular stop, and is different from the arrival time at a previous stop(i.e.row).
but for my other table BUS_ROUTE i have 2 varrays describing the links and stops enroute. in quite a few procedures i have to loop through these arrays and perform some calculations in every iteration is varray the best way 2 go, or nested tables?
Thank you
Amina
As an aside, external procedures tend by their very
nature to be slow - there's an overhead incurred
each time we step outside the database. Therefore
you really ought to avoid using a C extproc unless
your calculations really cannot be done in PL/SQL or
a Java Stored Procedure.
Also, before you go down the VARRAY route you should
consider the virues of analytic functions, notably
[url=http://download-west.oracle.com/docs/cd/B1050
1_01/server.920/a96540/functions56a.htm#83619]LAG()[/u
rl]. I think you really ought to do some
benchmarking of parformance before you start afdding
denormalised columns like ETA. You may find the
overhead in maintaining those columns exceeds their
perceived benefits.
Cheers, APC