IT Innovation Pays Dividends for Financial Leaders

By Thomas Kejser, Fusion-io
www.fusion-io.com 

Time is money.  In the finance industry, the truth of this adage has driven leaders to the forefront of the technology adoption curve.  When microseconds and milliseconds mean the difference between profits and losses, financial organisations soon build up teams of phenomenal IT experts to ensure the infrastructure powering the company and its services runs at top speeds.

Many of these teams focus entirely on managing huge datasets, and when it comes to processing massive amounts of data, time is often the first casualty.  For this reason, financial companies, who use big data to better understand their customers, their habits and preferences, are constantly striving to build faster, more powerful systems to meet the needs of their businesses.

From trading to risk management, financial applications perform more efficiently, when they run faster.  Many IT leaders are improving the performance in their large databases by using flash memory installed in data centre servers to deliver data to financial applications.  Using flash in the server, close to the CPU, removes the time applications spend waiting for data to travel across a network fabric to backend storage.

The shift to using storage class memory to accelerate applications can literally pay dividends for financial companies.  Let’s take a look at some examples of the results flash memory can deliver for financial IT.

Mega Scaled Transactions

 Credit card companies process enormous amounts of transactions daily from endless sources.  From retail locations to websites and now even mobile payment systems, tracking transactions is conducted on a massive scale.

To explore how flash memory could handle these enormous dataset sizes, last year, I tested how a single Microsoft SQL Server database could process 1.1 million transactions per second, logging one debit, one credit and one update of account per transaction.  The system scaled to process 25 billion simplified transactions per day from a single server.  To put that transaction number into perspective: this is equivalent to all the credit card transactions projected to be made daily by every individual on the planet in the year 2050, when it is expected that the world will be inhabited by nine billion people.  Such powerful single server scale challenges the conventional wisdom of using scale out systems for OLTP.

For financial organisations, ensuring that their systems can efficiently perform at these speeds can ensure they can scale with their own success.

Flash Powered Trading

The NYSE uses flash-powered servers in its Mahwah facilities to accelerate trading.  By adding terabytes of flash to servers, trades can be executed more quickly.  Flash also ensures performance needs can be met at busy times of the day or busy days in general, like when a hot stock conducts its initial public offering.

With billions of trades (not to mention dollars) handled every day, maintaining fast performance is key to ensuring investors don’t have to think twice about what happens on the backend when they click to submit a trade.  The ultimate goal of the flash fueled data centre is to keep systems running smoothly without any delays.  This goal can be a tough challenge to tackle for those still using disk-era architectures instead of flash memory in servers.

Big Data: Too Large For Disk

Using disk to handle large datasets means one of two things will happen.  The first outcome is delayed responses, something that is unacceptable in the financial industry as seconds can quickly add up to millions of dollars of opportunity lost.  The other option is to scale out infrastructure, adding massive arrays whose size is paralleled by increases in operating and capital costs.  As these systems are scaled out, the law of diminishing returns applies: Doubling the amount of arrays you have in your data centre does not guarantee you’ll double the amount of performance that trickles back to your servers.

Implementing flash as a memory, similar to DRAM, can combine capacity and performance into one convenient server-based acceleration package.  In addition, adding terabytes of flash memory to your server is much more affordable than adding or upgrading a million dollar SAN – often by up to 90% more affordable.  Those savings are compounded through reduced operating costs, such as power and cooling.

Flash in the enterprise is no longer viewed as an emerging technology.  With many of the world’s leading companies using flash to power their data centres, its safe to say that flash is fueling the future of financial computing. 

 

Comments

[Michael Rogosin]

There are hardware and software choices in building low-latency infrastructure. Flash and FPGA technologies are being used and developed. Computer languages (Java, C++, etc.) have their pros and cons. The choice of Java Virtual Machine can also make a huge difference - see www.azulsystems.com. Of course the best infrastructure with sloppy programming will lose to average infrastructure with great programming, but if you had both . . .

[Kisalay]

So developers can't write great code, let's fix the servers. I am not anti-Flash, just that it's creating a community of developers who are no longer bothering to get their basics right about performance being a core pillar of their development cycle.