Start-up software vendor CacheBox Inc. emerged from stealth this week with the release of Cache Advance, a bare metal caching layer that runs on top of storage and pushes flash-based acceleration closer to applications.
The inaugural version of Cache Advance includes "application-specific" modules (ASMs) for integration in MySQL and MongoDB workloads. The product is targeted at enterprise software developers who write custom storage applications for online transaction processing and online data analytics, including financial services organizations and multichannel retailers.
The software provides read and write caching and can accelerate applications on physical and virtual servers. The architecture includes a kernel I/O monitor that sets policies, an application I/O monitor for creating acceleration maps, and "flash-aware" logic for storage.
The Cache Advance ASM algorithm balances static and dynamic caching policies to target key application components, such as indexes and tables, and applies intelligence at the block level to manage input/output requests from high availability flash storage.
The static policies deal with information such as whether to use 4 KB or 8 KB flash caching, and deliver a package of preconfigured settings for application tuning. Dynamic caching analyzes an application's behavior and takes predictive actions on storage, such as prefetching the most frequently accessed data.
Cache Advance is engineered to run on commodity hardware that hosts virtual servers. Interoperability tests were performed on various vendors' hardware, including SanDisk's Fusion I-O, Kingston Technologies Corp., Micron Technology Inc., Seagate Technology PLC and Toshiba.
Cache Advance supports server-side cache and guest-level operating system cache for MySQL and MongoDB environments running either Microsoft Windows Server 2008 or Linux-based storage, including CentOS, Red Hat and Ubuntu.
The initial release supports the Kernel-based Virtual Machine hypervisor for Linux. The product roadmap includes a framework to support VMware vSphere hypervisor in 2015, which the two companies are expected to announce later this month. Support for Microsoft's Hyper-V is not included in the initial release.
The evolution of server-side caching technologies
Server-side flash caching technologies have evolved to address I/O performance issues, but they are limited to storage at the virtual machine or server level.
"Accelerated caching takes more than having high-performance storage. From our perspective, caching is best located very close to the applications. We're advancing the system code with our ASMs to enable a new paradigm of application-centric acceleration," CacheBox COO John Groff said.
CacheBox was founded in 2012 by technology executives with a background in architecting file storage systems. The privately funded company is preparing for its first outside funding round. CTO Dilip Rande and Vice President of Engineering Murali Nagaraj helped patent the "cache-coherent" Veritas Storage Foundation Cluster File System, now part of Symantec Corp.
Nagaraj said Cache Advance is designed to bridge the gap between hard disks and all-flash storage.
"In a large datacenter, you may have a small flash cache for restore of virtual machines that shares direct-attached flash. If you activate at the block level, you potentially could be caching the wrong type of data in flash. And we believe you have to look beyond file acceleration to see how the files are structured and consumed," Nagaraj said.
MySQL applications tested with Cache Advance were 12 to 20 times faster than an HDD-configured SAN, while OLAP applications in MongoDB were 110 times faster. CacheBox does not include an ASM for Oracle databases because "Oracle's Automatic Storage Management is already flash-aware," Nagaraj said.
"The key differentiator for CacheBox is that it's hardware-agnostic, whereas with most vendors you have to buy their PCIe cards," said Marc Staimer, president of consulting firm Dragon Slayer Consulting. "The algorithm is interesting in that it learns an application's behavior."
But while support for MySQL and MongoDB may be enough to get CacheBox started, Staimer said "they'll have to move quickly to support more applications" to gain broad enterprise adoption.
Cache Advance is priced on a per-CPU basis, with an annual licensing fee of $999, which includes the installation package, block-level application accelerators and ASMs.
CacheBox's roadmap also contains features for data deduplication and compression. Prototypes are being demonstrated with advanced beta testing expected to start in October, Groff said.
LSI Nytro PCIe cards get capacity, feature upgrades
HDS will build its own flash storage, sans acquisitions
NetApp upgrades high, low and all-flash FAS array systems