Nathan Moffitt

The Forgotten Flash Frontier?

Blog Post created by Nathan Moffitt Employee on Jan 28, 2016


Flash and the original high performance market: Mainframe.


Over the last several years there has been an explosion in the number of startups producing hybrid and all-flash platforms designed to accelerate business operations. All tout high IOPS, low latency response times and competitive costs, but how many of them are directing innovation towards mainframe, the segment that sets the benchmark for performance? Sadly, none.


Luckily for mainframe customers, HDS hasn’t forgotten them. In addition to our lineup of open systems flash offerings, we have resilient flash solutions for high speed, mission critical mainframe environments.


To get an inside look at flash tends with mainframe, we sat down with 2 gentlemen that live mainframe every day: Rick Szarama and Bill Smith.


Let’s start off with the basics, can mainframe customers benefit from flash:

Bill: Absolutely. Mainframe applications have similar needs to open systems applications. They have to process data - lots of data - and do it in a short periods of time. Flash helps by reducing latency to sub-milliseconds and increasing performance beyond what disk-based storage can provide.

Rick: Right. What’s important for mainframe users to understand though is that the benefits of flash don’t line up 100% with what open system applications will see. It’s close, but mainframe is different.

Talk a little bit more about that, how is the benefit of flash for mainframe different.

Rick: Most mainframe applications are extremely read intensive. Their read cache hits can easily be higher than 70%. On a well-designed mainframe with large cache sub-systems, most of the operations are served directly from memory. While flash is great for reads, large cache sub-systems are better.


As a result flash isn’t going to deliver the earth shattering results you see in the open-systems world. You get value, but not an exponential improvement. That has slowed flash adoption. 


Is that changing or is mainframe still a ‘disk drive first’ environment.

Bill: It’s absolutely changing. Flash, especially with offerings like the HDS flash module (FMD) is now cost competitive with high performance disk and we see flash adoption on mainframe accelerating.

Can you give us an example of the workloads you do see mainframe customers using with flash?


Bill: A lot of it depends on the customer and their industry. Processing of credit card transactions, stock trades and cell phone calls are areas we see flash used. Other applications using IBM DB2 and the WADS (write-ahead data set) in IBM’s IMS database are another good set examples.


Rick: And as said, as the price of flash has continued to drop, we’ve seen more customers are interested in considering flash as an alternative to 15K HDD.


How has this affected flash adoption in mainframe, especially all-flash array adoption?

Bill: Flash is being adopted, just over a longer timeline. Right now probably 30% of mainframe customers we’ve spoken with at multi-vendor roundtables (e.g. SHARE) are using flash. As they go through refresh cycles though they are absolutely looking at flash.

Near term that use is mostly hybrid but increasingly we see all-flash installations occurring and we are going to stay ahead of the curve there. In fact we have a few things in the works.

Rick: Another factor is resiliency.  Mainframes run extremely important operations and if they fail, the world crashes. That’s a big part of why we’ve seen adoption of HDS develop flash modules (FMD) versus SSDs.

Talk a little more about that. Why are they choosing FMDs over SSDs. Hitachi does offer both?

Bill: It isn’t that there is something wrong with SSDs, they are fine, it’s just that FMDs are better. HDS engineers FMDs specifically for uptime and resiliency, which is critical for mainframe. FMDs are also more tightly integrated with the our storage OS, SVOS. That gives enterprise mainframe customers greater peace of mind.


Rick: That part is huge. You can pull stats from SSDs and alert based on the information you receive, but because HDS manufacturers the FMDs and SVOS, there are tighter linkages that provide improved management and monitoring of flash wear levels as well as performance. This helps both us extend flash life and provide more predictable failure rates.

Does that give HDS a leg up in helping mainframe customers solve problems?

Bill: It gives us a huge leap. First off, we are one of the few providers in this space. Designing storage for mainframe isn’t easy and the resiliency requirements from these customers are demanding to say the least. That scares off start-ups and mid-range players.

Rick: Even some of the traditional players are having to re-think their strategy here. IBM offers flash on their DS8000, but at a much lower capacity and performance scaling than mainframe customers need. EMC is in this space too, but they also have flash scaling limitations around performance and replicating large consistency groups.


Bill: Is EMC even supporting mainframe on the VMAX?

Rick: Not yet. I think it was pushed out again? I’m pretty sure we are the only vendor offering a scalable, cost-effective all-flash storage solution for mainframes and we’ve got our support for LinuxONE.


At this point I had to cut the conversation because we started to get into competitive discussion, and while those are fun, they can stir up some trouble.


Net, using flash with mainframe can definitely add value, but it isn’t as cut and dry as on open systems. You have to take an extremely close look at resiliency, price and how the application will access the drives. Have more questions? Feel free to send us a message or visit the mainframe page for more details.