Software Development Life Cycle (SDLC) Case Study – Result = $440M Loss

Share this: [wpsr_retweet] [wpsr_plusone][wpsr_linkedin]

 

Software Development Life Cycle (SDLC)Importance

Solid Logic Technology’s foundershave experience across the financial industry and specifically in the development of quantative trading and investment systems. Many of the things we’ve learned along the way impact the way we develop software for clients across other industries. Most notably we’ve learned that software quality is extremely important and ‘software bugs’ cost lots of money. The study below shows how important in-depth software development, testing and launch management is for a company.

As early as August 1st, 2012 reports came out that Knight Capital Group,  a prominent electronic market-making firm specializing in NYSE equities, lost an estimated $440 million dollars due to a ‘software bug’. The news spread across financial news networks like Bloomberg, NY Times, CNBC and The Wall Street Journal.  Knight and other similar firms, trade US equities electronically using sophisticated computer algorithms with little to no human involvement in the process. While we will probably never hear the full story behind the ‘software bug’, it is suspected that a software coding error that was not quickly identified caused the loss. The loss is approximately 4 times their 2011 net income of $115m. It appears to have pretty much decimated the firm and at this point it looks like the firm will be bought or end up in bankruptcy.

While unfortunate, this example has some implications across any software project.

So what can we learn and take away from this incident?

  1. Software is not perfect, especially right after it is released
  2. A more comprehensive Software Development Life Cycle (SDLC) process and launch plan probably would have reduced the loss to a more reasonable amount.
  3. Always have a contingency plan for a new launch
  4. If new software is ‘acting funny’ then it probably has a problem and needs to be pulled from production and fixed
  5. When possible, conduct a series of small ‘pilots’ or ‘beta’ test along the way in a lower impact way
  6. If you cannot fully test the changes, then implement them slowly to minimize the potential errors in the beginning
  7. Have a ‘kill switch’ and know how to use it
  8. Have a formal SDLC process and follow it for all revisions
  9. Use source control for all software changes
  10. Have a defined launch process
  11. Have a way to quickly revert the changes implemented back to the previous version.

These are basic best practices that all software development firms should follow in order to consistently develop high quality software. Its unfortunate that there is a case study like this but these type of incidents are more common (but not to this scale) than most people imagine. I’m sure the group at Knight completed many of the above items, but something got away from them.

We put a huge amount of thought and effort into the process of software development and the consistent high level of quality that a solid process brings. We’re currently working on publishing a set of Software Development best practices – please contact us for a pre-release version.

More posts by Eric

[gravityform id=”4″ name=”Subscribe to our Blog” description=”false” ajax=”false”]

Share this: [wpsr_retweet] [wpsr_plusone][wpsr_linkedin]

Amazon EC2 Cloud Computing Cost Savings

This post is a long one and is part of an on-going series of some of the benefits we’ve identified in our experience in using Cloud Computing technologies, most notably Amazon Web Services (AWS) and different VMware products.

Overview

“The cloud”, specifically Amazon Web Services, has dramatically changed the landscape of High Performance Computing and Big Data Processing in recent years. Many things are computationally possible that would not have been a few short years ago. An organization can cost-effectively setup, launch and use a seemingly limitless amount of computing resources in minutes.

Most of the news media today is focused around using Hadoop on “Big Data”. SLTI has experience with this technology, but what happens if your task data set doesn’t fit nicely into this framework?? The writeup below is how we handled one such challenge.

Business Problem

The problem SLTI was trying to solve fits into the Business Intelligence/Data Mining area in the financial industry. The problem tested different inputs for an algorithm that is the basis for a quantitative equity trading system.
The algorithm had complex mathematical calculations and processing requirements across a large and diverse data set. The problem required testing a wide range of input parameters across four dimensions. The algorithm was tested across sixty-two different data sets. A summary of the size of the problem is shown to the right – We basically have to analyze 9.9 billion data points to come up with something actionable.

While the program logic is specific to the financial trading industry, it has many common concepts shared across different industries – engineering, legal services, etc. The question to ask is simple –

How many processing tasks have you had that you wished ran faster? Can they be split into multiple pieces and run in parallel? How can this be done cost-effectively?

Information Technology and Software Solution

Cloud Computing has dramatically changed the cost landscape of IT Infrastructure, especially for prototype or short run projects like this one. In a general sense, CPU cycles and RAM are cheap compared to the highly skilled labor required to improve performance by several orders of magnitude.

Our goal was simple – make the program to run a lot faster with minimal effort.

We have a large list of projects to be completed so development time is our most precious resource so we didn’t want to re-write the entire program. We kept the software changes and technology solution simple – it’s basically an 80/20 approach to setup the infrastructure and handle the code changes that still solves the problem, albeit in a less elegant fashion.

To accomplish our goal, we modified the program  to operate on a user-defined subset of the original data set. This allows the problem to be split into many small parts and spread apart across multiple servers. We can then distribute the pieces to each server to handle the processing for that subset.  

IT Infrastructure Architecture

In staying with a 80/20 simple solution first approach, we created a solution with the following pieces:

  1. Linux based application server (Amazon EC2 Amazon Machine Image (AMI), alternatively a VMware image could be created and converted to an AMI.
  2. Highly-Available, scalable, central filestore (Amazon S3)
  3. Master configuration data stored in Amazon S3

The cluster itself is comprised of sixteen cc2.8xlarge EC2 instances. Each instance has 88 Compute Units, has 2 x Intel Xeon E5-2670 processors (16 cores per instance), 60.5GB of RAM, 3370GB storage. The cluster provided 1408 Compute Units, 256 Cores and 968 GB of RAM.

The basic logic of the program goes something like this

  1. Load all required data into Amazon S3
  2. Launch the pre-configured AMI to run the program after the server launches
    • Get a specific subset of the data for the node from the central filestore
    • Update the master configuration data to notify the other nodes what data still needs to be processed before, during and after each test run.
    • Save the results to the central filestore
    • Shutdown the node after the work is completed

Cost Analysis

This is not intended to be a totally comprehensive cost comparison but rather a quick TCO comparison using some standard costs. To quickly do this, we used the AWS EC2 Cost Comparison Calculator on the bottom of this page.

SLTI’s EC2 based approach is roughly 99.5% cheaper than an in-house solution.  There are other similar examples of the ROI of an EC2 based approach for this type of workload here 

Key Takeaways

  1. Using the cloud enables a much more adaptive and scalable data processing infrastructure than in-house IT hardware.
  2. If you’re not using AWS (or something similar), you’re overpaying for IT infrastructure, especially for short run or highly variable workloads.

This post is a short overview on some of the ways we’re using advanced cloud computing technology to help our clients improve their IT agility and reduce IT expenses. We’re currently working on a few case studies that describe these concepts in more detail. To get updated with new research just sign up using the form on the right of this page.

If you’d like to explore a specific use case for your situation – please contact us

[gravityform id=”4″ name=”Subscribe to our Blog” description=”false” ajax=”false”]