Imagine what you could accomplish if you could do things 10X faster. If athletes could increase their performance by 10 times, they wouldn’t be concerned with finishing their workouts faster, they would be concerned about how many Olympic medals they could fit around their neck! What if you could apply that same performance to challenging problems today? You could realize the vision of improving people’s healthcare through faster life sciences analysis. Thanks to the wonders of science and brilliant people, scientists have figured out how to sequence DNA at incredible speeds. Today, scientists are capable of sequencing an organic sample of a cancer patient and then can compare that to tens of thousands of other patient DNA to identify the best possible treatment. An incredible achievement in science, but it can take a week or longer. What if we could do this in less than a day? This would mean faster time to treatment for the patient and 10 times the number of patients treated per year.

Today, many of those goals can finally be achieved thanks to the fastest storage solution ever built, the EMC DSSD D5, a Rack-Scale Flash solution for the most data-intensive challenges and real-time analytics.

When DSSD was founded in 2010 by some of the most brilliant systems and storage engineers on the planet, they were not focused on solving yesterday’s problems. Instead, they were focused on building a new storage solution that would not only solve the I/O intensive analytical challenges of today, but also make it future ready for the problems we will see in the years to come.

Over the past several years, customers have attempted to address their I/O challenges in other ways. One solution was to add flash to existing arrays or purchase an all flash array.

All Flash Array and Server Attached Flash

Rack-Scale Flash is different. It eliminates the compromises you make with traditional approaches. Rack-Scale Flash borrows the best from the All Flash Array and Server Attached Flash models to develop a solution that is dense and shared, which is flexible enough to work with both legacy and next generation applications, and combines unparalleled performance with enterprise reliability. Rack-Scale Flash is a quantum leap in storage that enables data and performance intensive applications to solve today’s complex problems as well as enables innovative new applications designed to solve tomorrow’s challenges.

DSSD D5 Performance Numbers

Let me recap what DSSD delivers across these dimensions.

In terms of performance, there are many innovations, but let’s review the high level around our Performance Oriented Architecture and discuss how DSSD D5 maximizes performance and optimizes the data and control path. DSSD delivers the world’s largest PCIe mesh fabric internal to the D5 enabling DSSD to deliver one massive 144TB pool of flash directly to the server. Every client has a direct connection to every flash module and all data movement is through the PCIe fabric and directly to the flash modules. The D5 controllers contain the processors which manage all storage attached to the DSSD appliance. DSSD’s Flash Management Stack runs in DRAM at the Controller level so it can reach across our Flash Modules. This is where we do things like defect management and global wear leveling. The typical flash management stack runs at the SSD level, meaning that individual SSD drives need to be stand-alone, have resource-constrained firmware, and a limited power budget—all of which results in slower and uneven performance. In addition, our controllers sit off to the side of the data path and set up and tear down I/O requests, but don’t handle the data directly. No slow store and forward architecture here.

DSSD Architecture - Connect to 48 clients

Another key focus area for DSSD has been to bypass all the layers evident in legacy stacks to get data to the application faster. So, when the application wants to do an I/O, it goes directly and efficiently to the app from DSSD without interruptions.

DSSD Architecture

In terms of Dense and Shared flash, this has also been a big focus. First, our Flash Modules are built from raw NAND and then tuned to get the best economics and differentiation at the media level. We’ve focused on the density of the packaging of our Flash Modules—over 18,000 NAND die operating in parallel to ensure the best power and cooling. Our Flash Module was built to burst higher and to be able to cool at higher performance levels. We can burst up to 50W compared to 25W in a server PCIe slot or SSD—twice as much power = twice as much performance.

Also, important was our focus on providing Native and Flexible Data Access. Today, applications store varying shapes and sizes of data, and then have to cram them into block-sized chunks. Most applications can convey more information than “read this block, write this block”—more information leads to efficient designs. Our interface with applications is a major differentiator. We natively connect to a variety of application types. We provide customers three options.

  1. Flood Block Driver interface (the world’s fastest) that allows customers to use legacy block device applications without the need to modify their existing application I/O source code in any way.
  2. Flood Direct Memory API. Any application can be modified, or new applications can be developed to use our Flood Direct Memory API. With our rich API, applications written to take advantage of DSSD storage have direct, fast access to many types of objects for easier and more efficient programming.
  3. Finally, Flood Plug-ins. Our mission has always been to have DSSD or our Partners leverage our rich APIs and build plug-ins to other common application interfaces or certain middleware layers. DSSD’s first plug-in was built in partnership with Cloudera: the DSSD Hadoop Plug-in. Finally, it is worth noting that all of these I/O modes can be active simultaneously across clients.

The final key attribute is Enterprise Reliability. DSSD has also been maniacally focused on delivering enhanced reliability. We have advanced global algorithms that help us provide consistent protection across the entire DSSD flash pool at all times versus data protection on a per SSD basis as in standard flash. Of course we’ve delivered all the key things you would expect: resilience of connectivity, HA controllers, data vaulting, enterprise grade ECC, etc. We have also delivered some very unique innovations. One of the most innovative is our patented multi-dimensional RAID algorithm we call Cubic RAID™. Other multi-dimensional RAIDs do not interlock the different dimensions because one dimension is implemented at the SSD drive level and another implemented by the array controller. Cubic RAID™ sees all of the flash, all of the time, and can optimize how data is protected.

A few months ago I was talking to a mentor of mine who retired from the tech industry several years ago. He was obviously amazed by what DSSD can do with storage performance today, but his simple question was what do you need all that performance for? After giving my standard speech on saving lives, building products faster, solving tough challenges that previously were unsolvable, I tried to break it down for him. It is all about Real-Time Analytics. Today, there are three categories of data and performance intensive workloads impacted by legacy storage bottlenecks. First, you have high performance DBs and DWs (think Oracle, MPP DBs, as well as some NoSQL and in-memory DBs). Second, High Performance Apps running on HDFS (like Hbase) and finally custom applications running on a variety of high performance file systems. With the changing nature of applications, customers require storage platforms that can support a combination of these key data workloads because each of these categories on their own may not be able to support an entire Analytical workflow, and organizations typically use a combination of the above. And, to add to the complexity, each of these data workloads has different performance profiles and constraints. So, customers leveraging legacy infrastructure are forced to create workarounds that are complex, underperforming and inefficient.

In future blogs, we’ll talk a lot more about the challenges we hear from customers around these workloads and how we help solve them.

Today, DSSD is helping customers do some amazing things. For instance, I/O intensive analysis and verification in the Electronic Design Automation (EDA) space can happen 10 times faster, which means technology companies can innovate faster. Companies, Governments, and Consumers who combined lose $5 trillion per year to fraud, can transition from trying to detect fraud, often days after it happens, to fraud prevention. We have customers using the DSSD D5 for many more things: delivering more comprehensive risk analytics in less time, improving the accuracy of predictive models, expanding the amount of data analyzed for real-time or streaming analytics, improving transaction models, simulations, genomics, etc.

The opportunities here are really endless and it is awesome to bring something to market that really is game changing for customers.

The Adobe Flash Player is required for video playback.
Get the latest Flash Player or Watch this video on YouTube.

Want to learn more? Check out the DSSD homepage and follow @EMCDSSD on Twitter for our latest announcements and information.

Matt McDonough

Matt McDonough

Sr. Director, Product Management and Marketing at EMC² DSSD
Matt McDonough

Latest posts by Matt McDonough (see all)

This entry passed through the Full-Text RSS service – if this is your content and you’re reading it on someone else’s site, please read the FAQ at fivefilters.org/content-only/faq.php#publishers.

Related Posts

Facebook Comments

Return to Top ▲Return to Top ▲