In-memory Heap Size of 2TB Attacks ‘Garbage Time’ Jitter in Java Applications
Targeting large in-memory Java computing implementations, Sunnyvale-based Azul Systems is releasing a new version of Zing, its scalable Java Virtual Machine for enterprise applications that includes a whopping in-memory heap size of 2TBs per Java instance, which Azul claims leads other Java runtime solutions.
Whether it’s real-time data analytics, high frequency trading or a travel reservations e-commerce platform, Howard Green, Azul’s vice president of marketing told EnterpriseTech, the typical reason Java applications freeze “isn’t the database, it isn’t the network, it isn’t the interface on the front end. It’s the Java runtime itself.
The problem is “garbage collection,” the process by which Java pauses to clean up its memory allocation, pauses that can take a few hundred milliseconds up to, in extreme cases, minutes. In the case of a trading platform, garbage collection jitter can result in failed trades; in the case of travel reservations, it can jeopardize purchases.
Green said Zing release 16.01 complements newer commodity cluster servers with up to 6 TB of real memory – only a few years ago a 64 GB server was considered large – which “means you can actually use all that memory right out of the gate with your existing Java application.”
“If you want to manage the equivalent of a very large amount of customer data or pricing data and hold it all in memory you’ve got a tremendous performance, quality of service and end user experience advantage over somebody else who’s relying on a database or something similar where there’s some gears and pulleys involved in the process,” Green said.
He added that competing JVMs, such as Oracle HotSpot, experience pauses at 12 GB or more of heap, requiring applications to be divided into smaller pieces. “Two TB is a long way from that,” he said.
Green said that many organizations are perfectly successful addressing the garbage collection problem with in-mem databases or external caching subsystems, “but they’re adding another piece of software, another layer of complexity to their application. We’re trying to make things in a lot of cases a whole lot simpler.”
“The Zing platform has provided an answer for Java users requiring large memory resources, low latency, and predictable and fast transaction performance for the past five years,” said John Abbott, Distinguished Analyst at 451 Research. “Now that multi-TB servers are readily available, Azul is making it possible to capitalize on the latest generation of hardware without requiring Java-based applications to be chopped up into small 4-12GB instances.”
Formed in 2002, Azul has more than 100 employees. Customers include Wells Fargo, Sungard, Credit Suisse, Priceline and Saks Fifth Avenue.
Azul said the Zing release also includes optimizations for Cassandra 3.0 with a 50 percent more efficient CRC32 implementation; new functions that reduce operating costs, including MXBean notifications that eliminate polling for key metrics; and support for a variety of Ubuntu and Suse versions of Linux.