Okay, on to CLSA. Here’s a base case on why CLSA/e Equity Access nixed
Sybase ASE 11.9 for Microsoft SQL Server 2000 on their research database….
Problem to solve:
Provide a mechanism that will enable secure electronic contribution and
storage of information related to equity research data, as well as, a
mechanism to enable secure electronic distribution of the information stored
in a repository or data warehouse. Factors affecting the implementation
* Providing access to the mechanisms via intranet and internet connectivity
* Ensuring the contribution / storage / notification process can be
performed in ; 5seconds once the information has been submitted for
* Ensuring performance (i.e., user response time) as it relates to the
distribution of information can be performed in ; 5 seconds once requested
* The contribution function must support no less than 1000 simultaneous
users located in 15 countries.
* The distribution mechanism must support not less than 2,000,000
simultaneous users located in 57 countries.
* Rapid startup time (3 months for a prototype – 9 months for production)
Those were my requirements…
Our parent firm was already using a home-grown research database for
contribution and distribution of 400 GB of research information. It was
suggested that we copy their setup and code to meet our rapid startup
Our parent firm had implemented a Sybase ASE 11.9 database running on
high-end HP Unix servers (hp superdome with 256GB memory with a 2000GB (2TB)
mirrored storage array)engineered for large-scale databases. This high-end
architecture is ideal for the specific demands of online transaction
processing workloads and high availability (never goes down). Unfortunately,
the whole setup is very expensive and very centralized (Hong Kong). In order
to meet their performance criteria, they resorted to distributing all or
parts of the database across multiple servers in different time zones. This
was very expensive solution to maintain because of the database licensing,
hardware, communication costs associated with distributing the data, and
personnel to support the distributed databases (3 databases x 55 tables in
each database x 4 countries).
We were starting with 6 countries and over 100,000 simultaneous users. The
parent solution cost would have an initial capital expenditure of $4,200,000
and a run rate of $46,000 per month (to start). Our budget to start was some
where around $125,000 (not including personnel).
This is how I make a living…
Using a copy of Microsoft SQL Server 2000, I prototyped the database and
supporting structures (indexes, stored procs, tables, etc) in two weeks
using our parent firm and another successful internet-based research
supplier as templates for our base service. The license for the software was
$900, and the server was already being used for other non-production work.
SQL Server license was free because we already had a development license.
baseline for our standards for development. Notepad was the development
tool. I picked up a free evaluation copy of ColdFusion 5.0 for the
middleware components and downloaded JDK 1.3 from Sun.com for the Java
The majority of the database activity revolves around a master transaction
table for collecting and distributing data. Seperate reference tables (in
1-to-many relationships with the transaction tables) are used for user
access, markets, countries, clients, contributors, names, address, etc, in
order to minimize table maintenance and complicated screen inputs.
Command-line batch jobs using SQL BULK INSERT we created to transfer data
between database servers.
In six weeks we were ready to test the system’s performance on a Compaq
Proliant 8500 class machine with 8 CPUs, 16GB memory, with external storage
for 200GB. We had