Three years after quietly launching, startup Datagres Technologies Inc. is preparing to bring its file analytics...
and caching software to full availability.
Datagres' PerfAccel host-based software is designed to find I/O patterns and hot areas within files to accelerate throughput and reduce latency for flash storage. "We sit on the server side, figure out what's going on and provide deep analytics around throughput hot files and so on," Datagres CEO Ranajit Nevatia said.
Nevatia said Datagres is preparing a new version of the software with a new GUI, greater file-level analysis of I/O patterns and write-back caching to go with write-through support in the first version. Palo Alto, California-based Datagres originally pushed PerfAccel as a performance tool, "but the best part of the technology was the analytics," Nevatia said. "So we're repositioning as analytics."
The biggest change in PerfAccel since its 2012 soft launch is it is now licensed separately as PerfAccel Analytics and PerfAccel Caching. The analytics application looks at usage patterns of network data across the server, network and storage. It measures IOPS, latency, throughput and bandwidth at file, folder, node and grid levels, and identifies hot data and read/write hits and misses. It identifies issues that cause latency, and predicts performance and capacity problems.
"By monitoring IOPS, we can tell how the underlying storage is performing," Nevatia said. "Admins can schedule jobs and provide appropriate resources to applications so they don't run into those problems."
The caching mode is built on top of the analytics engine and improves performance by distributing data to appropriate tiers. It moves IOPS to the server to separate performance from capacity.
Nevatia said common use cases for early customers include applications with I/O performance challenges such as NoSQL databases and larger rendering farms. The current version of PerfAccel is designed for large-scale Linux environments, runs inside KVM and Xen hypervisors and supports Docker containers. The product does not yet support Windows or VMware.
Ranajit Nevatia, Datagres CEO
Nevatia said most early customers start with 10 to 50 nodes, although the software can scale to 5,000 nodes. Datagres' largest customer is up to 900 nodes connected to a NAS device on the back end.
"The Webscale guys are trying to move IOPS to the server, but they have an army of engineers doing that," Nevatia said. "We are moving in the same line -- your server is where you want your IOPS happening. We are making the servers more powerful, and getting IOPS done more efficiently."
Pricing is an annual subscription model, based on the number of nodes supported. Nevatia said "high volume" customers pay $1,000 per node per year.
Nevatia also said vendors such as Virtual Instruments on the analytics side and PernixData and Infinio for performance can be considered competitors, but those vendors focus mainly on VMware virtual machines and block storage.
Ashish Nadkarni, IDC analyst for enterprise servers and storage, said Datagres can also serve as an alternative to NAS acceleration appliances sold by Avere and others. He said Datagres' analytics gives it insights into more than just the storage array, and he thinks the startup can do well with organizations with I/O issues.
"You could buy an expensive solution like Avere for relief, but you're off-loading the issue to the network," Nadkarni said. "Datagres knows what the host is up to, it knows what the storage is up to, and it knows what the network is up to."
Nadkarni said solving the problem on the host is a better method for improving performance than doing it on the array. "People need to start thinking in a different way," he said. "There's a mentality that's been pushed by storage vendors that all problems need to be resolved at the array level. They don't want to deal with the host. That mentality has to be reversed, because there is so much happening on the host."
Nevatia took over as CEO this year from founder Srinivasan Viswanathan, who remains with the company as CTO and chairman. Nevatia previously held marketing executive positions at Riverbed Technology, Panzura and Bitcasa.
Big data puts extra demands on storage
Making the case for data-aware storage
Using tiering, caching to speed up performance