Schubert Ticketing Services
Conducting Sales Harmoniously
Selling tickets for theater, opera and sporting events might seem like basic
transaction processing. But the core of this business is seat analysis,
which rivals decision support in complexity. Here's a sample request: "I
need the best available orchestra seats for a party of 10 for Phantom
of the Opera on any Wednesday night over the next two months, except
the third Wednesday of the month. One member of our party is in a wheelchair
and will need an aisle seat." Two other factors add to the complexity.
Dozens of similar queries can hit the system at once, and the definition
of "best available seating" changes constantly.
To meet this kind of demand over and over, New York City-based Schubert
Ticketing Services, which sells tickets to most Broadway and Off-Broadway
productions and numerous other events, has adopted a parallel database on
parallel processing hardware. "What parallel processing can do for
us is gather up lots of seats simultaneously--if you have your data partitioned
properly," says David Andrews, director of operations and chief technology
executive. "A database that can support parallel updates can fire off
multiple data requests that can be handled in parallel. We will do as many
things in parallel as possible so we can get one request out of the way
and the next one under way."
Schubert Ticketing chose a clustered SMP architecture of Digital Equipment
Corp. servers running Digital Unix with Informix's OnLine Dynamic Server
as the database engine. Clients will be 486 PCs running Windows 95. "The
data is centralized; the clients are all over the place in six states,"
says Andrews. The firm opted for the RISC Alpha chips because they support
64-bit processing, and only DEC was ready with its 64-bit architecture on
Schubert's timetable. "Informix's partitioning scheme meant superior
performance," Andrews adds.
The database engine for the ticketing application is just part of a $7 million
migration to a client/server system. But ticketing--the mission-critical
business process--is driving everything else. Schubert has more than 600
PC clients accessing the database from dozens of outlets in six states.
It sells about 11.5 million tickets--worth about $600 million--per year.
Andrews is leading his 35-member IS staff through a two-year migration.
"Today we're on the ragged edge. I wouldn't want to deploy it,"
he says. The application is still under development, and Andrews says it
will need all the performance that the Alpha and OnLine Dynamic Server will
offer by the time he's ready to deploy. "When we spec'd what we needed
a year ago, that's where we are on the performance curve today, but we want
50 percent above that for our safety margin," he says. "That will
be here a year from now."
Schubert presently uses a Control Data mainframe with a proprietary database.
"It's finely honed and tuned to ticketing but in a single-threaded
manner," Andrews explains. "So we are dependent on a single CPU
for our analysis." There is enormous administrative overhead in this
system. "Every time you make a change in the database, you have to
write a program to restructure all the in-memory data pointers."
Although plans aren't final yet, he anticipates using six to eight Alphas
in three separate boxes. As a hedge against growth, each box can handle
as many as 12 CPUs. Andrews plans on 1GB of RAM with a 100GB database. "We
typically have about 20 million-plus seats available for sales at any one
time and a customer phone database of about three million entries per year.
Today we're limited in historical data analysis, and it's not cost-effective."
That will change. Andrews envisions a data warehouse for decision support
using extra processors.
"To get the power out of SMP you have to be cognizant of your business
activity and its data," he advises. "Then be proactive about how
you partition your database and how you spread it across CPUs and disk drives,
in order to maximize the performance of the database."
The company has not brought in many consultants, because the IS development
staff of 10 understands its data and how people use it. "Customer-service-oriented
solutions are becoming more important. You want someone to pick up the phone
in 20 seconds and get you on your way quickly," Andrews says. "IT
decisions have to flow from the business process. For Schubert, everything
is driven from ticket sales."