Welcome to our blog! Welcome to our blog!
New Mainframe Workloads and Cognitive Computing Powered by All-Flash Arrays

All-Flash Arrays Enable New Mainframe Workloads and Cognitive Computing

This post is an update of an article that originally appeared on Dancing Dinosaur.

A few years back flash storage cost so much that technology gurus advised organizations to use flash selectively. Sure, flash delivered impressive input/out per second (IOPS) performance, but you could only afford to use it for the most critical performance- or throughput-constrained applications. Most recently, IBM began recommending flash for cognitive computing, too.

Today, high performance flash is widely available, and between its vastly improved economies of scale and recent technology advances, the cost of all-flash enterprise storage has dropped substantially. IBM boasts an entry-level all-flash array starting at $90,000. This changes not only the way you think about storage, but also the way you think about data center performance and the workloads you run, including cognitive computing.

And it’s not just IBM. Most enterprise storage vendors are introducing all-flash storage. Driving this interest is skyrocketing demand for high performance and real-time analytics. The volume of enterprise data is growing exponentially, and the use cases for it are expanding just as fast.

Business managers are demanding real-time analytics, predictive analytics, compliance monitoring, fraud prevention, and more. With fast flash storage and a high performance mainframe, you could potentially stop a fraudulent transaction while the perpetrator was still at the teller window or cash register.

For cognitive computing, mainframe speed matters

For Cognitive Computing, Speed Matters

As described by IBM in the announcement of its three new all-flash mainframe storage products, the company envisions these devices for more than the z System’s core OLTP workloads. According to the company, they are designed to provide the speed and reliability needed for workloads ranging from enterprise resource planning (ERP) and financial transactions (OLTP), to cognitive applications like machine learning, predictive analytics, and natural language processing.

Related: Mainframe Meets Machine Learning

IBM did not just slap the usual flash cards into these boxes. Rather, the company reports undertaking a complete redesign of the flash-to-z interaction. As IBM puts it: through deep integration between the flash and the z, IBM has embedded software that facilitates data protection, remote replication, and optimization for mid-range and large enterprises.

The resulting new microcode is ideal for cognitive workloads on z and Power Systems requiring the highest availability and system reliability possible. In short, they will be fast enough to catch bad guys before they leave the cash register or teller window.

Download this free eBook to see how mainframe computing is now critical to Big Data.

To further drive speed, IBM reports bypassing the device adapter to connect the z directly to the high performance storage controller. IBM’s goal: reduce latency and optimize all-flash storage – not just affect a simple replacement by swapping new flash for ordinary flash or HDD.

“We optimized the data path,” explains Jeff Barber, IBM systems VP for HE Storage BLE (DS8, DP&R and SAN). To that end, IBM switched from a 1u to a 4u enclosure, runs on shared-nothing clusters, and boosted throughput performance. The resulting storage “does database better than anyone; we can run real-time analytics.” Barber further states that the typical analytics system – a shared system running Hadoop – won’t even come close to this level of performance.

Mainframe Performance: Results Matter

“High performance and reliability have led us to continually deploy newer DS8000 models as new features and functions provided us new opportunities,” says Bojan Fele, CIO of Health Insurance Institute of Slovenia. “Furthermore,” he continues, “[DS8000] has increased employee productivity, ensuring we can better serve our clients.”

These kinds of results are sure to increase interest from lines of business (LOB) and marketing departments for real-time access to mainframe data (SMF, RMF, log data, and more) for business analytics. Could you blame them?

You can also learn about key mainframe optimization challenges, opportunities and use cases for 2017, spanning DB2 and network management on z/OS, as well as new ways to save on your monthly IBM MLC charges and new options for long-standing mainframe issues via Syncsort’s webcast.

Alan Radding

Authored by Alan Radding

You can find Radding's work in a variety of leading publications and online Web sites. Major research and consulting firms also use his research and writing services. In addition, he writes for a variety of corporate clients. You've probably been reading his technology and business writing for years without realizing it; much of it is done as a ghostwriter. View his work at technologywriter.com
0 comments

Leave a Comment

*