In-Memory Computing for All (Ft. Intel Optane)

Dale Kim | Nov 5, 2019

If you had a choice of processing data in-memory versus not in-memory, all other things being equal, wouldn’t you always choose in-memory? You might not need the higher performance, but if it were available to you, you’d take it because faster is always better than slower, right?

It would be wonderful if we lived in a world where the fastest computer medium was easily accessible to all. Everyone would be using that medium. Certainly, in-memory computing has gained a lot of momentum in recent years, but the fact is that RAM still comes at a relatively high cost which makes it less than practical as the primary medium for data storage. While the costs of RAM have dropped dramatically over the years, the amount of data we manage has grown even more dramatically. We currently have reasonably priced laptops with 16GB of RAM that probably could’ve stored a medium-sized company’s entire data set 30 years ago, but our data sets have grown so much more in that time. We need another medium that can store that expanded amount of data. And never mind that RAM loses its data when the power is shut off, which limits its usefulness in some situations where resilience is critical.

It’s safe to assume that in the near future, nobody will erase the gap between the speed of RAM, and the capacity and cost-effectiveness of disk drives and SSDs. Though as the eternal optimist, I do believe that one day there will be that universal medium that is the fastest, most scalable, and cheapest. This might sound silly now, but come back to this blog in 300 years and we’ll see who’s laughing.

But jumping back to the present day, we’re very excited about the upcoming release of Hazelcast IMDG because it is certified to take advantage of the speed, density, and cost advantages of Intel® Optane™ DC Persistent Memory. The lower cost of Optane versus RAM, with performance characteristics fairly close to RAM, gives more organizations the opportunities to gain competitive advantage. For example, retailers can now expand their ability to handle high transaction loads, including during peak demand events.  Telecommunications companies can run complex calculations more quickly to make network optimization decisions in real-time to maintain a positive customer experience. Banks can run multiple machine learning algorithms within milliseconds to quickly analyze transactions and deliver accurate approval decisions.

It’s important to note that there are two separate modes for Optane, which means Hazelcast customers are able to tailor the deployment to the requirements of the application. The first mode makes Optane look like RAM, so it’s almost as fast as typical RAM modules, but much more economical and dense in terms of capacity. Any business that previously thought the speed of in-memory was hard to justify now has an answer. They can get near RAM speeds at a lower price, and now move workloads that benefit from higher speeds. At the same time, this Optane mode is volatile, so like RAM, the data gets lost when the power goes out. While this seems to contradict the “Persistent” in “Optane DC Persistent Memory,” there also is a second mode that enables persistence so that data is retained even when the power is shut off.

In persistent mode, Hazelcast can use Optane as a fast-medium for the Hot Restart Store feature. In other words, Optane acts as a faster SSD for non-volatile storage. This mode is much slower than the volatile mode of Optane, but when persistent memory is the use case, the advantage over SSDs is very clear. This feature lets Hazelcast store the contents of its memory allocations to a persistent store, which can then be used to repopulate the in-memory store quickly once a node is restarted after maintenance activities. Since many Hazelcast use cases deal with many transactions and computations per second, any amount of downtime can be troublesome. With Optane, you can minimize that downtime with a faster restart capability, so you can get on with business. For a closer look at the performance of Hazelcast on Intel Optane, some initial benchmarks are available in the white paper, Advancements in High Speed, In-Memory Systems, which shows that restarts can be up to 3.5x faster with Optane.

As we continue fine-tuning our optimizations, we will run benchmarks to show you Intel Optane and Hazelcast in action. If you’ve had prior reservations on the use of in-memory technologies due to cost, we’re hoping this new innovation will encourage you to revisit this topic soon to explore the use cases for your business or application which benefit from in-memory computing. Stay tuned for more information on this growing trend.

About the Author

Dale Kim

Sr. Director, Technical Solutions

Dale Kim is the Senior Director of Technical Solutions at Hazelcast and is responsible for product and go-to-market strategy for the in-memory computing platform. His background includes technical and management roles at IT companies in areas such as relational databases, search, content management, NoSQL, Hadoop/Spark, and big data analytics. Dale holds an MBA from Santa Clara, and a BA in computer science from Berkeley.