The general appeal of appliances in financial services boils down to:
-
Higher performance – Performance may mean lower latency, higher throughput, more consistent behavior, or a combination of the three. Appliances often include specialized hardware that eliminates performance bottlenecks to accelerate repetitive operations.
-
Consolidation/cost reduction – Often a single appliance can do the work of many servers running software, resulting in a smaller datacenter footprint, less power consumed, fewer servers and software licenses and ultimately, lower costs. Whether trying to avoid costly colocation facility fees for front office solutions, or simply reducing datacenter costs in the back office, this is a focus of virtually every kind of financial firm.
-
Simplified operations – Appliances offer turnkey installations that software cannot, usually shipping with everything installed and ready to work out of the box. With software, you have to install the server, install the operating system, patch the operating system to whatever level is supported by the software license, then install the software, tune and test it. There is a multiplier effect (O/S versions times software versions times server-to-appliance ratio) that can make it 5 to 20 times harder to maintain and operate a scaled software solution than equivalent appliances.
In financial services, the degree to which each of these three areas of value matters to you depends on what problem you’re trying to solve, but generally speaking you can think of the value of appliances in terms of front-office vs. back-office requirements.
Front-Office: The Need for Speed
If you’re building front office applications, time is money, and performance and profits are measured in microseconds. This leads buyers of electronic trading components to obsess about low latency, and just as importantly, consistent latency. The increasing uptake of hardware appliances as a means of accelerating front office operations is well documented with products like the Solace Message Router, TIBCO Messaging Appliance, Exegy Ticker Plant, and XtremeData dBX. And those shops most serious about latency install monitoring appliances (like the Tip-Off product from TS-Associates) so they can effectively monitor all their other hardware accelerated appliances.
Most of the successful front office appliances have embraced specialized hardware such as FPGAs or network processors, because just packaging software on a server and calling it an appliance doesn’t offer a big jump in performance. Specialized hardware also minimizes jitter by eliminating the OS issues that can plague high-volume low-latency software deployments.
Back-Office: Easy Does It
In the back office, buying priorities shift. Of course performance still matters, but back-office applications are generally more about reasonable performance with very high reliability and a guarantee of delivery. In the software world, it takes a lot of effort to achieve horizontal scaling with the redundancy it takes to ensure 24×7 operations. Well designed appliances reduce the number of managed devices, and come with fault-tolerance and failover built in.
Appliances often also provide operational visibility that spans both the hardware and software, making it easier to understand behavior and thresholds for operational planning.
In the back office, you will find a wider range of appliances since the application set is so much more diverse. Banking customers are some of the biggest buyers of some of the products mentioned in the Information Week article, such as the Oracle Exadata and IBM Netezza. Beyond that, IBM WebSphere Datapower and Layer 7 XML Gateways provide SOA and web services solutions, F5 offers a range of security, storage and load balancing appliances for retail banking, and Solace offers a high-performance guaranteed messaging appliance for accelerated transactions and WAN distribution.
Whatever the appliance, or its role, the key to succeeding in the back office is in reducing costs, improving reliability and making life easy for the operations teams.
Financial Services: Leading the Move to Appliances
As is often the case with technological change, financial firms are leading the charge, but it really is different strokes for different folks. Few industries can match the performance demands of front office trading, and few operations teams have to deal with more complicated operations in the back office. Where the need exists, the solutions emerge, and appliances are forging a permanent home within the financial services IT landscape.
(This article originally ran on Larry Neumann's blog, which can be found here.)
Innovations in Trading and Technology (
Comments | Post a Comment
3 Comments to "What Does the "Rise of the Appliance" Mean for Financial Services?":
kmcpartland
12 October 2010
I agree with nearly everything you say. The ROI based on the reduction in needed data center space and related power - going from 10 racks to 2u - makes many of these products sell themselves before the performance issue even comes up. However, on the front office side for the lowest latency traders I've heard that appliances are less than ideal as they add an extra hop to the process. One way or another data must come in and out of that appliance for trades to be made. These same people are more enamored with card based solutions that can slot right into the same server housing the black box algo. Clearly there are many nuances here - some appliances can house trading algos, for example - and the number of trading firms with the latency sensitivity to make the above matter is relatively tiny, however those same firms are also the ones with the bleeding edge ideas that ultimately trickle down to everyone else. Will appliances ultimately shrink down to cards? Some already have.
Comments (80)
larryneumann
14 October 2010
Kevin, appliances have their place, but aren’t the answer to every problem. Some solutions are well suited to software, some to acceleration cards, and some to appliances. Which solution is best for which situation comes down to the specific problem and objectives. The article I referenced as the basis of the article was focused on Oracle and IBM (and others) moving towards appliances for mainstream computing, and my intent was to illustrate how appliances are affecting the mainstream of financial services firms. Some of my example were front-office focused and some were back office focused. The thing we have to be careful of is not to associate hardware and appliances in financial services too tightly with front office. It’s true that if you sell latency-differentiated feed handlers, you're probably selling primarily to high speed traders. But in the bigger picture, the high-frequency trading segment is a small-niche of financial services, and a small portion of the target for appliances overall. For example, in our business, front office trading represents about 10% of the total messaging market, and the microsecond-sensitive world of high-frequency trading is less than half of that. When we deploy an enterprise deal for a big investment bank, the vast majority of messaging-centric applications are in the mid or back office, which is where the “shrinking the datacenter while improving manageability” value sells well. The most common architecture we are seeing for latency obsessed hedge funds or prop traders is to load their feeds, algos, pre-trade risk, etc. into a single many-core server. Yes, it gets rid of the appliance, but it also gets rid of the network replacing the idea of hops anywhere with shared memory exchanges between cores. We have an IPC/shared memory messaging protocol that does pub sub between cores with latency of about 400 nanoseconds. Inevitably though, that one server with 16 cores turns into two, then three, then four servers. What is the right architecture then? Do you rebuy the feeds and quadruple your software licensing costs, or introduce a network and messaging take the performance hit of a few microseconds? Again, it depends on your latency and cost tolerance. Another example of it depends: Increasingly, algorithms are based on arbitrage or hedging strategies across multiple markets, which means WAN communications between the trading centers becomes the performance bottleneck. What difference does shaving 5-10 microseconds in New York mean if it takes 40 milliseconds to move data to London? The total solution can be much faster if the WAN is made 20% more efficient using hardware compression (as an example). To your question on appliances and acceleration cards, it seems that, outside of the front office, the market is reducing to blade servers for general software execution and appliances where the hardware/software cost reduction or management advantages justify. You can’t really put third party cards into blade servers, which means you would be introducing a rack server if you want to consider acceleration cards. It makes just as much sense to introduce an appliance, given it is probably the same number of RUs and a similar amount of power, etc. Thanks for the comment, it's an interesting topic!
Comments (6)
larryneumann
14 October 2010
I just learned that comments can't have line breaks. Gak! You''ll have to imagine where my paragraphs begin and end :-)
Comments (6)