Qumulo Inc. came out of stealth today with what it describes as data-aware scale-out NAS with real-time analytics...
Qumulo Core was designed by many of the same developers who created Isilon scale-out NAS. Qumulo founders Peter Godman, Aaron Passey and Neal Fachan were responsible for dozens of Isilon patents. EMC acquired Isilon for $2.25 billion in 2010.
CEO Godman said that while Seattle-based Qumulo waited until today to officially launch Core, it has had paying customers using the system since late last year. Qumulo has been issuing software updates every two weeks to its early adopters, mostly in media and entertainment.
Qumulo Core is a software application that runs on top of Linux on a hybrid flash storage system. Godman said Qumulo offers Core as a software-only product, but most customers have bought it pre-packaged in a Qumulo Q0626 appliance that uses commodity hardware. Each 1U node includes four 6 TB Helium hard disk drives and two 800 GB enterprise multi-layer cell solid-state drives (SSDs) for 25.6 TB of raw capacity.
The nodes use Intel Xeon E5 1650 3.5 GHz CPUs and 64 GB of memory. Each node includes two 10 Gigabit Ethernet ports. Customers can cluster from four to 1,000 nodes.
Qumulo's goal is to scale data, not just files
Godman said that after interviewing more than 600 storage end users, Qumulo founders determined the main problem with NAS today is scaling data rather than just files. Qumulo then set out to give customers information about their data that was previously not available.
"Managing data at scale is largely about answering questions," he said. "What do I actually have, how fast is it growing, who's using it, who's not using it, what data never gets used, what can I archive, what do I need to back up, where does the performance go, what applications are consuming it?"
Godman said Qumulo built a database into the file system to answer those questions. He said the goal of Core is to combine NetApp's performance for small files with Isilon's handling of large files and easy scalability.
The SSDs store Core's file system database. Godman said Qumulo qualifies hardware platforms for customers who only want to buy Core software, ensuring they have enough flash to run efficiently.
He said Core is designed to be so easy that "anyone in your company should be able to use [it]."
The system's real-time analytics dashboard shows information such as the number of nodes in a cluster, how much data is being used, how much is available, and the IOPS and throughput the cluster generated over the past 24 hours. Users can drill down into each directory to see how much capacity it is using and how frequently each file in the directory is being accessed.
Ant Farm, which creates content for movie studios, television networks and video game publishers, began using a four-node Core cluster last September, according to IT director Sam Frankiel. He said Core storage helps Ant Farm scale for seasonal work, such as the June Electronic Entertainment Expo (E3).
"We're now in the middle of three to four months of intense work for video game advertising for E3," Frankiel said. "For E3, we go from 10 artists to around 40, and we double the number of our render files. The load hitting that type of storage was slowing us down. We needed something faster. Now we're hitting more spindles for more speed."
Frankiel said Ant Farm evaluated NetApp and Isilon before picking Qumulo. He said the biggest differences in Qumulo's favor were price and the visibility it provides.
"We can see at a glance which disks are getting hit during peak times, and we can shift files automatically during those times," he said. "The visibility it gives us into our storage usage is something we haven't seen in any other product."
Sportvision, which creates graphics such as the yellow virtual first-down line for NFL broadcasts and Pitchf/x that tracks every pitch in a baseball game, purchased an eight-node system with 200 TB of raw capacity in January. IT manager Grant Turner said Sportvision works with a lot of high-definition video and generates up to 4 TB of data for one baseball game when Pitchf/x is used.
Turner said he has been impressed with Core's analytics engine.
"Their analytics are way better than any other storage that I've seen," he said. "We use it for data lifecycle management. We can tell what data is hot, what's being actively accessed, what we've used a lot and what hasn't been touched in weeks. Then we migrate data off to tape or nearline storage accordingly."
Data-aware: A new trend?
Qumulo isn't the first vendor to use the term data-aware storage. DataGravity also uses that description for the Discovery Series it began shipping in August 2014. So far, the two newcomers have concentrated on different uses cases. DataGravity is aimed more at compliance and data classification, while Qumulo is tackling media files. But the vendors are looking to prove that you don't need object storage to manage large-scale file storage.
"Conventional wisdom has been that the only way to deal with one billion files was with object-based storage," said Arun Taneja, consulting analyst at the Taneja Group. "Qumulo just broke that mold and destroyed the myth that file systems cannot deal with billions of objects."
Enterprise Strategy Group analyst Scott Sinclair also sees Qumulo scaling beyond traditional NAS because of its data-aware and analytical features.
"Object storage came about on the idea that even scale-out file systems have limits," he said. "Qumulo is weaving in more advanced data elements to scale-out file systems much larger than they could scale before. Core is intended to store lots of content."
DataGravity announces data-aware storage arrays
EMC upgrades Isilon for Hadoop analytics