Success Story
Insurance Data Warehouse - ViPS, Inc.
(published August 1998, DM Review)
By Jerry Strausberg, ViPS Systems Architect
Background
MCSource is a data warehouse, analytic engine, and suite of business
applications designed for medical and financial decision-makers.
MCSource gives managed care organizations the ability to evaluate
and understand their data to improve quality of care, operational
efficiencies, and the strength of their provider networks.
ViPS relies on CoSORT for high volume processing in our
Data Integration (DI) stage. The DI process takes the 'Reformat'
client legacy data and performs various C, SQL, and other processing,
including large flat file sorts with CoSORT's sort control language
(SortCL). The data is now ready for fast, high-volume
loads through Red Brick's Parallel Table Maintenance Utility (PTMU)
into various Red Brick tables. After DI is complete, business analysts
can use the MCSource GUI for analysis and decision-making. For intermediate
testing and development, SortCL can change record
layouts faster and easier than a C or COBOL module can.
Platforms
ViPS Healthcare Information Solution's MCSource™ product
uses the Red Brick Data Warehouse and CoSORT Sort/Merge/Report,
Version 6.2, primarily on Sun UltraEnterprise servers running on
Solaris 2.6. Support for the HP 9000 and IBM RS/6000 are now implemented
as planned.
Problem Solved
When I joined the team, we were using a product called Op-Tech Sort,
which provided acceptable response time for files under 25MB, but
unacceptable response times for files larger than 25MB. At that
time, ViPS was negotiating an MCSource contract with the
Blue Cross and Blue Shield Association Federal Employee Program
(FEP), the nation's largest health plan.
A typical FEP file might contain 5 million records, exceeding our
25MB limitation by over 3 gig. VIPS quickly realized that sorting
the large volume of data would take nearly forever. In addition,
Sun's Solaris 2.5.1 was limited to 2Gb files. We needed the 64-bit
Solaris 2.6 version to address FEP's file sizes and a sort product
that addressed large file I/O directly. CoSORT's 6.2 version did
so immediately, while SyncSort had to suggest UNIX pipes and other
workaround commands. The result was that CoSORT's wall time was
40% of SyncSort's. On our development machine, a Sun UE 3000 (4
CPUs), CoSORT sorted a 7.9Gb file on 5 keys in 32 minutes.
Product Functionality
CoSORT (for coroutine sort) is a package of sorting-related modules
and tools for UNIX (and NT): 1) mainframe (MVS JCL) sort parm conversion
program 2) mainframe-based sort control language (SortCL)
program 3) UNIX and Windows NT system sort verb replacements 4)
Micro Focus COBOL and SAS sort verb replacements 5) Interactive
prompting (batch spec builder) program 6) API for C, COBOL and FORTRAN
sub and co-routine calls.
ViPS uses CoSORT's sort control language (SortCL)
to describe field-level data layouts and sort its clients' legacy
data. SortCL has an extensive set of intra- and inter-field
manipulation capabilities, including conditional selection and data
statements, cross-calculations, data type translation (EBCDIC to
ASCII, for example) and aggregation (summarization). It can all
be done in one same sort/report pass. The other product modules
interface to the central SMP sort engine.
Strengths
CoSORT has many interface choices, which helped us identify the
most efficient way to get our processing done. We were impressed
with all the tuning we could do to prioritize sorting on our multi-job
environment, and with CoSORT's overall price-performance ratio.
Weaknesses
Because we are familiar with mainframe sort products, we found CoSORT's
sort control language almost identical. We wanted to see additional
formatting features, such as the ability to specify a fixed number
of occurrences of a /DATA character and more traditional empty input/output
file handling. Both of these features were added in late 1998/early
1999.
Selection Criteria
After Op-Tech and the sorts inside UNIX, we evaluated two major
commercial sort packages, CoSORT and SyncSort. The major reasons
we selected CoSORT were performance, especially on large sorts,
functionality and technical support. Pricing was a significant plus
as well.
Vendor Support
If I had a problem, and there weren't many, the CoSORT people really
jumped on it. From a business standpoint, we were able to customize
an integrated licensing model that made sense to both ViPS
and its clientele's management.
Documentation
CoSORT comes with a well-organized, easy-to-read manual covering
each interface with tips and examples. On-line man pages, the .pdf
chapters, and statistical/warning/error display were helpful.
|
