Azul Systems, Inc. announced the launch of Zing Platform Edition with WebSphere. Zing PE integrates IBM WebSphere Applications Server (WAS) with an enhanced version of its Zing Java Virtual Machine.
InfoQ interviewed Azul CEO Scott Sellers about the launch and about Azul Systems.
InfoQ: Can you tell us what this announcement means for development teams?
Scott: Zing is a 100% Java compatible off the shelf JVM, just like HotSpot. So it can run anything a JVM can, application servers, JVM languages like Scala and Groovy, and so on. The Zing PE announcement is special in that we have introduced specific optimizations for WebSphere Application Server (WAS) that make it really easy to deploy Zing in the context of WAS deployments, which are traditionally complex, especially when deployed to multiple instances. So now teams can launch Zing PE which embeds the WAS server, and makes deployment a breeze.
InfoQ: So what does Zing PE mean for application owners running WebSphere-based workloads today?
Scott: Zing PE, like the Zing JVM itself, is optimized for mission- and revenue-critical Java applications running on current-generation Linux servers. We have released Zing PE to enable WebSphere-based systems to deliver performance without the latency jitter and outliers that can characterize some Java-based systems. We make it easier to configure the JVM by reducing the need for specialized tuning, we make it easier for applications to take advantage of large in-memory data stores, and in short, we help our customers meet their SLAs.
InfoQ: Why was WAS selected, was that based on market share?
Scott: That was a response to customer demand. As a relatively small company Zing offerings are largely driven by our customers. The one thing we have seen across the WAS landscape are mission critical deployments, portal applications, ecommerce apps.
Customers need performance consistency and lower latency. For example in electronic trading, the absolute lowest latency is a requirement. But even in apps with human interactions, response times needed to handle 100 ms worst case response times. So the requirement is different, but the need exists across the spectrum, and the off-the-shelf JVM's can't deliver that. so we have seen more mission critical apps with an emphasis on performance, and we have responded to that demand.
InfoQ: Will you be targeting other platforms as well?
Scott: The same things can be done with other app servers and other significant frameworks. As an example, frameworks such as Hadoop could stand for some optimizations, also in-memory caches and analytics engines, so this is the beginning of a longer term strategy to target specific platforms. Applications can always do better if we look at the runtime characteristics of the platform itself, and that's what we're doing with the WAS PE launch.
InfoQ: How do you handle Garbage Collection (GC)?
Scott: We have a completely novel GC algorithm called C4 (Continuously Concurrent Compacting Collector). Azul is the only vendor to ever fundamentally address GC as an orthogonal concern to size of heap and allocation rate. All other collectors have a characteristic that the world must stop; Azul is the only vendor that doesn't have to stop the world ever.
InfoQ: It sounds like the GC algorithm is a key part of Azul's success?
Scott: GC is a complex problem, and the algorithm is complex. So yes, it is one of our crown jewels, and we have patents, but we also have the knowledge and the experience of building it and making it work. That's something that took us years to get right.
InfoQ: Azul started life as a hardware company, but now you are strictly software. Can you talk about that transition?
Scott: Back in 2002 we recognized that for large scale business-critical runtimes, Java wasn't going to cut it. It wasn't like today. There wasn't any hardware infrastructure around to build a better HotSpot. So we started building hardware. We started shipping in 2005. We built three generations of products that included the microprocessors, software, everything. Around 2008, we spent time with chipmakers like Intel and AMD, and combined with their experience and new capacities of commodity platforms, we decided to become software only. Zing started shipping in 2010. But the Most important event in that transition was in the fall of 2011. We introduced a version that didn't require our proxy architecture. When we were a hardware company, we offloaded processes to their own context. The first version of zing had that architecture as well. But with Zing 5 in 2011, things got interesting, we did away with that proxy architecture and got to target the low latency market. A remarkable state of the market has evolved. There have been open source JVM's, Project Harmony, and through time and Oracle acquisitions, there remain only two vendors that provide JVM's Oracle and Azul. It is a vast market and there's very little competition. So we are in an interesting spot.
InfoQ: Can you talk about your pricing model?
Scott: Traditionally Zing has been sold on a subscription model. WAS PE is different. Regular Zing is subscription based, which allows customers to make a decision, and if not pleased, they can cancel. (Last year we had over 100% renewals. How can we be over 100%? Customer renewed for longer terms than before. So Zing is sticky, once in production, people keep renewing.) IBM prices WAS on a perpetual model based on Processor Value Units (PVUs). Depending on the type of processor and number of cores, IBM creates a PVU rating for a micro processor. So our pricing is roughly a percentage of that.
Some financial companies have said they will continue buying Zing as long as they make more money. They always buy it. It is easy to compare performance of apps on HotSpot and Zing, so the proof is always clear.
Zing Platform Edition with WebSphere is available immediately. It currently supports WebSphere Application Server ND, versions 7.0, 8.0, and 8.5 and is supported on Linux-based servers running Red Hat Enterprise Linux 5/6, CentOS 5/6, SUSE Linux Enterprise Server 11 SP1/SP2, and Ubuntu 10.04 and 12.04 LTS.