The big conference held in New York City in June of this year was QCon, and when we saw Gil Tene's name on the roster of speakers talking about how transactional memory is changing the way enterprises are provisioning their server hardware, we decided to get in touch with the Azul CTO and put together a podcast on the topic.
If you're not familiar with them, Azul is the company that's famous for scaling the JVM to over 700 cores way back in 2007, so they know a thing or two about optimizing server hardware transactional memory and scaling systems into the stratosphere. So what's Tene's take on the integration of hardware transactional memory (HTM) into Intel's popular line of i3, i5 and i7 chips? "It's a pretty exciting thing when you have a lot of cores," Gil Tene, CTO and founder of Azul Systems, says. And given the fact that a quad-core processor has become commonplace, even in the world of handheld devices, configuring enterprise server hardware with a small army of Intel chips isn't unusual at all.
Out-of-the-box performance improvements
We're going to learn over the next couple of years what HTM is good for and what HTM isn't good for.
Gil TeneAzul Systems CTO
Of course, HTM won't turn a sloth of a system into a cheetah. The impact the enhancement will have has a strong dependency on the type of applications the hardware serves. Applications with fine-grained access to data won't likely see large benefits, while applications that do a significant amount of list processing in which locked data structures cause contention between running processes stand to benefit the most. Much of this magic happens at the hardware level, which means programs don't necessarily need to be rewritten to take advantage of this new feature. "HTM has the potential to improve performance and execution of multithreaded code without people having to do much about it," Tene says.
And while existing systems need not be rewritten to take advantage of HTM, there are indeed situations where systems can be optimized knowing that these features are available. So there is an unknown variable, which is the question about how software developers might change the way they develop APIs, code operating systems, optimize containers like the Docker, host virtual machines and even perform basic list-based operations knowing that the HTM feature will be available to them at runtime. "We're going to learn over the next couple of years what HTM is good for and what HTM isn't good for," Tene says.
To hear the full conversation, and even to find out what's new with Azul's Zing JVM, listen to the full interview between Azul's Gene Tene and Cameron McKenzie in the accompanying podcast.
