LAS VEGAS -- Claiming that block-based storage caching has been commoditized, startup Datagres Technologies Inc....
on Tuesday previewed the production version of its PerfAccel Analytics Platform, which adds granular application-level I/O analytics to the vendor's file system analytic software.
Datagres formally launched in 2012, but is just now bringing its product to full availability. The startup showed off the enhancements during a demonstration at the NetApp Insight partner and user conference.
PerfAccel 3.0 consists of two products: PerfAccel Caching and PerfAccel Analytics, each licensed separately. Earlier versions bundled the analytics and caching on a single platform.
It includes Application Analytics that ties an application's performance to individual CPU processes or user session IDs. The Application Analytics can pinpoint hot spots within multi-node storage clusters. A feature known as Analytic Mode identifies problematic applications and predicts the impact on performance of adding flash-based caching or additional storage.
"It's a design tool to help a DBA [database administrator] or a storage admin figure out which flash storage hardware they need to buy next," Datagres president and CEO Ranajit Nevatia said.
Version 3 features an HTML5 graphical user interface that provides a single management pane, including centralized policy-based storage management across thousands of nodes. The upgrade adds support for Apache Cassandra databases, Red Hat 7 and Ubuntu 14 platforms, XFS file system, and most distributions of the AutoFS file systems. Version 3 also supports OpenStack as well as workloads running in KVM and Xen hypervisors.
PerfAccel connects application, storage layer for file-level 'intelligence'
Datagres joins a crowded field of vendors that sell storage caching software. Whereas most vendors cache data at the block or virtual machine level, Nevatia said his company provides "application-level intelligence" that works with any vendor's underlying storage. PerfAccel resides on a host server to examine application I/O access patterns at the file system level. The aim is to help customers pinpoint bottlenecks and boost the performance of underlying storage.
"We create a connection from the application to the storage layer. We sit close to the application and measure each and every operation of the file system. We will very accurately tell you, in a large grid, what your applications are doing and what's causing any performance or latency issues," Nevatia said.
Ashish Nadkarni, IDC storage research director
IDC storage research director Ashish Nadkarni said the convergence of servers and storage gives Datagres a wide field on which to play.
"It marks a shift away from monolithic storage arrays to host-level dynamic data distribution," he said. "What Datagres is trying to do is push data intelligence to the host, and then optimize I/O to streamline performance."
File-level caching to be rebranded
Datagres accelerates application performance by caching a copy of a file in flash, leaving the original data copy in primary storage. Instead of caching on entire copy of the file, PerfAccel caches part of the file or file metadata.
Nevatia said the caching label doesn't accurately portray how PerfAccel performs.
"We are trying to figure out what to call it," he said. "Caching really doesn't do it justice, because caching means you have to read the file to be able to place it somewhere. What we do is policy-based data distribution. We know which types of files will be needed to support a process, and promote those files from backend slow storage on hard disk drives to PCI Express-based SSDs fast storage."
The PerfAccel platform has been used by beta customers for about two years, Nevatia said. He said the number of users is in double digits, but declined to identify any customers.
Datagres sells PerfAccel as a yearly subscription license based on the number of nodes. Pricing starts at about $1,000 per node.
Datagres' initial focus is on grid-scale storage workloads for big data, high-performance computing and virtual data centers running Docker containers and KVM and Xen hypervisors. Nevatia said support for VMware and Microsoft Hyper-V virtualized storage are on the product roadmap, possibly in 2016.
Why we need data-aware storage
What's the difference between tiering and caching?
Learn the basics of flash caching and its popular options