Technology trends tend to mesmerize us, as we like to follow the latest and greatest trajectories. Software-defined storage, where storage management software is decoupled from the physical storage itself, is one such trend. Another is the adoption of flash for primary storage. Both are important and significant, but does that portend the end of traditional controller-based storage with a heavy dose of hard disk drives?
INFINIDAT would argue to the contrary, at least for enterprise-class storage. The company argues that its new controller-based storage architecture provides the performance, robustness, reliability and scalability required from enterprise-class storage and, moreover, can do so at a reasonable price.
Storage customers still suffer from the usual pain points, especially as the demand for storage seems to continue growing uncontrollably. Operational issues include performance and reliability; one can never seem to get enough of each. And as the storage demands increase, environmental issues can take center stage, such as floor space and power consumption. Last, but not least, is cost: How can enterprises afford to keep up with the costs? INFINIDAT believes it has the answer to these and other pressing questions.
The INFINIDAT storage architecture
The InfiniBox storage system instantiates the INFIDIDAT storage architecture, which is designed to address those challenges. This architecture is the brainchild of Moshe Yanai and his storage engineering team. Yanai is the storage guru responsible for the Symmetrix architecture that launched EMC’s massive growth spurt in the 1990s and continues to this day as the market’s leading enterprise-class storage system. Yanai is also responsible for the XIV storage system, which IBM acquired in 2008 and which has been commercially successful since then. In other words, when Moshe Yanai designs a storage architecture, pay attention.
Among the key architectural issues that storage professionals face are how to improve performance, reliability and scalability. Storage in an INFINIDAT InfiniBox storage system is composed of RAM, flash (SSD) cache and hard disk drives (HDDs). Now, except for protocol drivers, storage caches and HDDs do not need to be aware of hosts and volumes. The InfiniBox uses that point to advantage, so a standard system represents a collection of up to 3.125 billion 64KB objects that INFINIDAT calls “sections.” Every section has a checksum (which is used to check for the possibility of errors that may have occurred during the transmission of data) and what is called an Activity Vector (AV) score appended to it.
Collectively, the InfiniBox AVs form a heuristic map of inferred relationships between sections. The AV map is really a heat map (where usage of data indicates temperature) that enables the “hot” working set of data to be kept in high performance cache as the application workloads run. Serving all reads from DRAM and SSD cache (which are a relatively small portion of overall storage) achieves maximal performance without having to store everything on high performance (and high cost) memory and storage devices. As a result, InfiniBox can support high performance and high capacity while at the same time not breaking the storage budget bank.
Another key storage architectural principle is to utilize drive resources efficiently. InfiniBox achieves this by dispersing volumes and file systems across all system disk drives which yields statistically uniform drive utilization. Two key benefits from this approach are no hotspots (which would reduce performance) and fast rebuilds (which are essential in highly scalable systems).
Technically, the InfiniBox system takes 14 sections that have similar AVs, as well as 2 computed parity sections (for reliability purposes) to create a 16-way stripe. The system writes the 16-way stripe on 16 different drives selected from the disk drive pool using a non-deterministic algorithm. This means that there is no one hot disk whose failure can severely impact overall performance and availability. Plus, rebuilds are fast since if one disk fails the rebuilding process continues on the remaining 15 drives working in harmony for the one stripe rather than the smaller number used in standard storage architectures, but overall for all stripes all drives in the system participate in the rebuild and not just 15.
Among the InfiniBox system’s other architectural characteristics are triple-redundant power and data paths, 3 active-active-active nodes (usually one active-active node is typical of most storage systems, so this goes beyond that), all nodes connected to all drives, and data verification is performed end-to-end. Those are among the reasons for INFINIDAT’s assertion of 99.99999% uptime (downtime of no more than 3 seconds per year).
A single InfiniBox system can support up to 480 high capacity (6TBs) HDDs, enabling up to 2PB of usable capacity in a single 42U high rack. That enables customers to capture what INFINIDAT asserts is an average 6:1 floor space consolidation. In fact, density, power management and floor space consolidation are among the factors that lead INFINIDAT to claim that InfiniBox typically delivers 65% to 80% cost reductions over existing enterprise-class storage solutions.
In addition, InfiniBox offers core data management services, including snapshots and asynchronous replication, for all protocols. Those protocols cover SAN, NAS, and the mainframe.
The net result is that the INFINIDAT storage architecture is available in a commercial solution, the InfiniBox storage system that meets essential enterprise-class storage user requirements for performance, reliability, scalability and cost.
To paraphrase Mark Twain, the reports of the death of controller-enabled enterprise-class storage systems have been greatly exaggerated. Enterprise-class storage users demand a stringent combination of requirements, including performance, reliability and scalability, but at reasonable costs that often inhibit the acquisition of leading-edge, high-performance systems.
INFINIDAT would argue that its storage architecture, as instantiated in its InfiniBox storage system, solves that problem. Not only can it compete effectively with existing enterprise-class storage systems, but the company also makes a strong case that neither software-defined storage using commodity hardware (which is not optimized for performance among other things) nor flash storage (which does not have the cost characteristics at scale among other things) can hold up to what the company offers enterprise-class storage customers.
From what we’ve seen of Moshe Yanai’s latest architectural innovations, enterprises struggling with the usual storage pain points would do well to investigate INFINIDAT’s InfiniBox storage solutions.
© 2016 Mesabi Group. All rights reserved.