In financial services, new technology is usually considered
a cost center. The battle between IT professionals, who want to deliver
leading edge systems and management, with its concern for the bottom line,
is familiar to all. But what IT professionals know, but often can't demonstrate
that technology redesign can have tremendous impact an the bottom fine
- has been proven at CS First Boston.
CSFB reported a record $108 million profits in the first quarter
this year, bolstered by a substantial increase in revenue from Fixed Income
Trading operations. These results came shortly after the firm's Fixed Income
Trading Technology group (FITT) completed a major overhaul of its trading
database support systems, working with UNIF/X Inc., a New York City-based
firm specializing in the design and automation of database management and
support systems.
FITT serves all fixed income trading departments for CSFB's
U.S. operations. The group is responsible for collecting and maintaining
data from distributed trading operations, and ensuring its reliability.
When Steve Allen, a vice president of FITT and the senior person
responsible for trading database support, came to work at CSFB in 1994,
he encountered a number of significant technical challenges. These problems,
like many industry-wide, lay with the firm's distributed database support
systems.
When distributed client/server systems emerged as the foundation
of next generation computing, CSFB downsized to Sun Sparcstations running
Sun O/S. However, to leverage previous investments, CSFB, like many Wall
Street firms, connected client/server systems to existing mainframes. Individual
managers added hardware and software from different vendors as needed.
Unfortunately, these pieces often didn't fit together as easily as the
vendor's marketing materials promised. The complexity of financial databases
also increased exponentially, as new financial products continuously developed.
"Because [CSFB] was relatively unstandardized, problems
required much more time to fix," explains Sergey Fradkov, senior technical
specialist at UNIF/X, who brought his experience working with distributed
databases systems to the CSFB project, "And because things weren't
configured properly in some cases, the system would crash more often."
Managing large database systems at the heart of the world's
financial markets has never been a simple task. But maintaining and supporting
very large relational databases in complex, heterogeneous environments
now consumes a substantial part of every financial company's technology
resources.
Surprisingly, trading database support groups are usually several
times larger than a company's trading system development group. Before
the revamp, CSFB had eight database administrators (DBAs), some of whom
were expensive outside consultants. At an industry average of about $100
hour, Fradkov estimates that CSFB was paying about $1.5 million a year
for support in FITT alone. Now the system is kept running with half the
staff.
When looking at improving productivity, and cutting back on
staffing, Allen found it important to have UNIF/X input. "I wanted
[FITT] people to think about how to eliminate their jobs," says Allen,
"which is not easy to get any employee to do, That's another reason
why I relied heavily on Sergey."
PIECEMEAL CONFIGURATION
FITT's database support systems were plagued by a variety
of problems, mostly relating to the fact that system components had been
assembled piecemeal, over time. The nonstandard hardware and software configurations
kept most maintenance and administration functions from being automated,
and many processes, including data downloads and system backups, needed
constant baby-sitting, unable to run successfully without manual intervention.
Support costs increased steadily in tandem with the growing size and complexity
of CSFB's databases - a tall order for any system support group.
System problems cut deeply into productivity. DBAs who took
turns answering calls during the "nightcycle" were woken to troubleshoot
so often they couldn't be productive during the day. And other DBAs spent
more than two hours each morning investigating more problems from the previous
night. Additional DBAs were brought in as consultants at a cost of almost
a quarter of a million dollars a year each. Though the group's size continually
increased, overwork remained a constant.
The company realized that these difficulties directly affected
the updating and access of databases worth billions of dollars. As a first
step to tackle the challenge, Allen and Fradkov analyzed existing hardware
and software configurations, and developed a comprehensive plan to implement
standards across all FITT systems. To save on maintenance and support costs,
all database servers were set up to use standard directory structures.
They also standardized disk and kernel configurations, as well as versions
of database software.
The next step was to automate and increase the reliability
of database maintenance applications, to minimize the problems occurring
daily and during the nightly download of data from back office systems.
Fradkov and Allen developed a sophisticated system, controlled by an Autosys
job scheduler, to completely automate database administration tasks. The
new system reduced the number of problems, as well as costly nighttime
support staff calls, by more than 85 percent.
Even after implementation of this, however, daytime DBAs still
needed to find out if any issues from the night before remained unresolved.
This involved a costly and time consuming manual double-check.
Allen worked with Fradkov to create a system to automatically
diagnose and report those system difficulties. "We found it would
be faster to develop software ourselves," says Fradkov. "We looked
at third party software, like Patrol, and it required quite a bit of work
to fit it into CSFB's environment."
Now, when DBAs begin work in the morning, they simply check
their email for detailed messages about problems they need to troubleshoot.
There is no need to launch a daily in-depth investigation to find the remaining
issues.
To eliminate the need to baby-sit database backup processes,
Fradkov recommended purchase of SQL Backtrack from Data Tools, which offers
advanced data compression technology to reduce miltigigabyte databases
to a size that is within a specified Unix file system's limits. "Data
compression improves the stability of system backups across databases because
it reduces the chance of running out of disk space while back-ups are running,"
Fradkov explains. "As a result, database maintenance applications
also are less likely to fail."
While overhauling FITT's database support technology, Allen
and Fradkov lived by this credo: utilities inherent in all systems should
be generic and dynamic. A generic utility should perform many different
functions. Dynamic utilities should be customizable with minimal effort.
They created a number of generic and dynamic utilities that
can be used for multiple support and administration tasks. For example,
a "permission manager" developed by the pair allows the business
units to determine which users should have access to applications and data.
"The best way for a DBA group not to be a bottleneck is to remove
themselves where their involvement is not really appropriate," Allen
says.
The impact of the technology changes made at FITT is ongoing.
New systems have reduced the number of staff and consultant DBAs needed
to support the Fixed Income Trading operation by more than 50 percent.
By eliminating the most expensive consultants, and using less senior staff,
FITT is saving over $1 million a year on headcount alone. Because systems
are up and running regularly, DBA support turnaround time is lower. In
effect, this "uptime" has boosted productivity - saving the company
even more money. The remaining database support group can actually solve
more problems for users, in less time.
Inside FITT, the work life of DBAs has dramatically improved
and morale is higher. "No more sleepless nights for DBAS," notes
Allen. "The redness is out of their eyes and the smiles are back on
their faces." Record profits aren't hurting, either.
by Brad Mehl Brad Mehl is a freelance writer in New York City.
|