• We’ve gone from a few million to more than a billion calculations processed in a 24-hour period and are saving tens of millions of dollars.”
• Business Unit VP at Tier 1 U.S. Bank

In this page: The Challenge | GigaSpaces XAP Delivers | How GigaSpaces XAP Does It | GigaSpaces XAP Track Record | Details for the Performance/Utilization Benchmark

The Challenge

Most IT organizations are focused on improving resource utilization, consolidating servers and saving costs, but many are unaware that one of the biggest factors affecting utilization is application performance. If an application is running at sub-optimal performance, it is in fact utilizing too many resources (CPU, memory, data center facilities, etc). Boosting application performance can mean running at the same speed using less resources, or in other words, improving utilization. For example, one GigaSpaces customer, eBay subsidiary Markplaats, is using GigaSpaces XAP to boost performance, allowing it to run the same loads with 95% less servers (read more below).

The root cause of most performance problems is the architecture. Traditional software architectures are highly centralized and complex, and thus prone to bottlenecks, which limit performance. The centralized database is the #1 performance issue; next in line are the large number of “moving parts” and the network hops between these different parts. These bottlenecks exist in all enterprise applications – not just on the high-end. And they cost your IT organization money across the board, by slowing down applications and causing them to run on much more hardware than is actually necessary.

GigaSpaces XAP Delivers

GigaSpaces XAP powers the world’s fastest applications, and is able to bring the same extreme performance to any enterprise application. Applications running on XAP become faster and more efficient, allowing them to meet the same loads with less hardware – meaning you can consolidate servers and save costs. To prove this, GigaSpaces hired independent consultants to perform a benchmark – they deployed an identical reference application on two commodity machines, one with a traditional architecture and a Java EE application server, and one with XAP. Both applications were resilient, highly available and transactional. The benchmark results are below:

(*) To ensure resiliency and high availability, every transaction was written immediately to the database. No data caching was used.

(**) The application used XAP as an application server, In-Memory Data Grid, messaging server and deployment framework.

At the bottom of this page are more details about the benchmark used to obtain these figures.

Additional Financial Benefits

Want more details about the financial benefits to your organization?

Let us know and a GigaSpaces expert will be in touch to help you quantify the benefits of boosting performance for improved utilization.

How GigaSpaces XAP Does It

• Single Platform – guaranteed end-to-end performance with no network hops
• Most enterprise applications rely on numerous pieces of middleware. GigaSpaces XAP unites all these different parts into one platform, providing all the middleware functions in-memory, in-process. This means that there are no remote network calls between different components of the application – XAP eliminates "network hops" which are a substantial barrier to performance. And because XAP manages the entire middleware stack, it is able to guarantee end-to-end performance from the user's perspective, delivering minimal latency from the load balancer to the database.
• High Performance – extreme performance and utilization for any enterprise application
• GigaSpaces XAP powers the world’s fastest applications – transaction processing at leading stock exchanges, real-time currency trading, and online gaming – and brings the same extreme performance to diverse fields including Web 2.0, ecommerce, telecom and back-office analytics. It does this by performing fast, in-memory processing using a top-of-the-line In-Memory Data Grid, which is as reliable as a traditional database without the performance penalty. XAP is also the only product that eliminates moving parts, resulting in zero network hops (see SINGLE PLATFORM above)
• Scalable On-Demand – scales up dynamically to deliver as much throughput as needed
• GigaSpaces XAP addresses the root cause of performance issues, the architecture. It does this by packaging the entire application, from the load balancer on the front end to the database on the back end, into a lightweight scalable unit, running entirely in-memory. Each unit is hyper-efficient, providing the minimal possible latency and the highest possible throughput on the given hardware resources. To improve performance, you simply parallelize the effort by duplicating this scalable unit on more machines. Best of all, if loads increase or hit an unexpected peak, XAP will automatically deploy more instances of your software on more machines to provide higher performance and more throughput.

Learn more about GigaSpaces XAP…

GigaSpaces XAP Track Record

GigaSpaces XAP has a long history of boosting application performance and improving hardware utilization. Here are just a few of our war stories and benchmarks.

Customer Stories

• eBay Subsidiary Marktplaats – Markplaats is a classifieds service that serves over 6.4 million unique visitors per month. It is using XAP to release data bottlenecks and parallelize processing, resulting in improved performance that will allow them to meet the same loads with 95% less servers. The server consolidation ratio is an unheard-of of 18:1. According to Cees de Groot, lead architect at Marktplaats: "The transaction speed XAP delivered for our sponsor ad engine is phenomenal … With these speeds, we expect that we'll need only a handful of machines to handle our current load … down from several hundred servers in our current data center."
• Tier 1 US Bank – uses XAP to power a risk management application that runs 23 hours a day on more than 10,000 processors, executing hundreds of millions of calculations involving terabytes of data. The Business Unit VP in charge said that “we’ve gone from a few million to more than a billion calculations … Computations for a single trade have been reduced from 4 hours to 40 minutes … an analyst running a scenario would have his data the next day. Now, a user … can have it 20 minutes later, and we’re running 600 scenarios a day.” This performance boost has enabled the bank to make better use of existing production resources and defer investments in additional IT infrastructure: according to the Business Unit VP, “the increased utilization … will save the bank tens of millions of dollars in IT costs over the next three years.”
• Patsystems – Patsystems provides high-performance e-trading and exchange systems to organizations such as Citigroup, JPMorgan, Morgan Stanley and Mitsubishi Corporation. It uses XAP to power its next-generation e-trading system, with very tough requirements for low-latency and fault-tolerance, while supporting reliable connectivity with exchanges around the world. John Mansfield, Global Trading Director, said that Patsystems benefits from XAP’s ability "to synchronize trading activities across geographies with extremely low-latency, " and added that XAP facilitates “a large volume of high speed transactions to meet the evolving needs of the trading marketplace and enable global control, 24-hour operation and exceptional resiliency."

All Customers

Benchmarks

In addition to the JEE performance benchmark referenced above, see the following performance-related benchmarks.

• Massive throughput for Java and .NET – GigaSpaces XAP demonstrated over 1 million read operations/sec. (1K payload) for both Java and .NET, on a single Sun4450 machine with Intel 7460 CPU and no special tuning.
• Low latency benchmark – XAP demonstrated 0.29ms latency for a remote call with an object containing 8 fields. For the same object, with synchronous replication to an in-memory backup, latency was only 0.69ms.
• High performance on multi-core – XAP demonstrated the ability to increase throughput up to 1,400,000 operations/sec., on a Sun T5240 Sparc machine.
• Web application performance – conducted on the classic ecommerce application example, Pet Clinic, running on commodity hardware and MySQL. XAP achieved 1.4 billion page views per day, with latency of only 6 ms.

All Benchmarks

Details for the Performance/Utilization Benchmark

Here are the basic details of the benchmark we used to obtain the hardware savings figures above:

• The reference application – a stock trading application with two workflow steps, validation and matching, performed on a feed of buy and sell requests. Domain model consists of an order, a trade and a stock symbol.
• The reference hardware – one Dual Core Intel(R) Xeon(R) X5260 CPU @ 3.33GHz, 64 bit, 4GB memory, with RH Linux.
• One application, two middleware implementations – XAP enables transparent migration for applications using traditional architectures. This made it possible to run exactly the same code with a traditional architecture vs. XAP.
• The traditional architecture (without XAP) – in this scenario the application ran on a popular JEE application server, using the app server’s JMS message broker, and a MySQL database. To achieve reliability and high availability, all transactions were committed synchronously to the database using two-phase commit. No data caching was used.
• The XAP-based architecture – in this scenario the application used GigaSpaces XAP as application server, message broker, and In-Memory Data Grid (which is inherently transactional). Resiliency was achieved by keeping in-memory backups of all data using the In-Memory Data Grid. All trades were persisted to a MySQL database in the background, ensuring 100% data consistency.
• Benchmark procedure – the application was wired to the traditional architecture, using a Spring Framework configuration file, and then deployed on the reference hardware and allowed to process a large volume of data. Throughput and average latency was measured. Then another copy of the same application was wired to GigaSpaces XAP using another Spring configuration file, deployed on the same hardware, and measured for throughput and latency. The results are the difference in performance between the two deployments.
• How did we calculate 80% hardware savings? The benchmark showed a 500% performance improvement. This could mean delivering five times more throughput on the same hardware, OR it could mean delivering the same throughput with 1/5 the hardware, an 80% saving. For a distributed application, this assumes linear down-scalability, which is provided by XAP.

For more details, see the benchmark page