Analyst: Simon Robinson
Sector: Storage & Systems
Report: type Impact Report
Date: Thu, 23 Oct 2003
The road to the development of successful high-performance, next-generation file systems may be littered with failures, but that hasn't deterred startup Panasas. The company is emerging from stealth mode this week with its ActiveScale Storage Cluster, an object-based approach to storage clustering that it claims offers up to 30 times the performance of any existing mainstream approach. Optimized for large-scale Linux clusters, Panasas says the system took over 230 man-years of development effort. But the company backs up its claims with more than just bravado; one of its first customers, Los Alamos National Laboratory, has signed what it says is the largest-ever storage capacity contract for a Linux cluster, agreeing to purchase up to 500TB of storage over the next year in a deal worth roughly $2.6m.
Impact assessment
Core message
Storage startup Panasas has emerged from stealth mode with a radical offering aimed at meeting the storage requirements of large Linux clusters. It also claims a significant customer in the form of Los Alamos National Labs.
Competitive landscape
Although in theory Panasas' approach puts it into competition with just about every NAS and RAID vendor, in reality it needs to convince early adopters of the merits of its object-based design.
The451 assessment
Panasas has taken a big idea and lots of R&D, and wrapped it up into a pretty compelling proposition. Adding support for CIFS and NFS is a crucial first step to acceptance "as others have learned to their detriment" and a keen focus on early target markets should provide the company with a solid foundation over the next year or so. The big challenge is whether it can take ActiveScale and object-based storage into mainstream computing.
Context: Panasas was formed in April 1999 by Garth Gibson, a prolific storage technology scientist with a particular focus on RAID, NAS, disk arrays and parallel and distributed file systems. Gibson created the Parallel Data Laboratory at Carnegie Mellon University. Panasas recruited Rod Schrock, former CEO of Alta Vista and head of Compaq's consumer product division, as president and CEO in early 2001. Senior VP of engineering Paul Mitilas has held senior positions at Ciena, ONI Systems and Nortel Networks, and coinvented optical power management in an optical network.
Panasas has raised $72.5m to date in three funding rounds, the latest of which was worth $32m, completed in September 2002. Investors include Centennial Ventures, Intel Capital, Novak-Biddle Venture Partners, Carlyle Venture Partners, Evercore Ventures and Mohr, Davidow Ventures. The company says it has enough money to take it through to profitability, which it expects to reach during 2005. The company currently has around 125 employees, the vast majority of which work in engineering.
Strategy: The company claims that existing storage networking architectures struggle to cope with 'scale-out' Linux-based compute clusters, which frequently comprise hundreds or even thousands of nodes, each one of which requires substantial bandwidth for handling high-performance applications. Although SANs deliver high bandwidth, they serve only a modest number of clients, while NAS systems can serve a large client population, but not at high I/O rates. Panasas adds that these clusters present a new management and reliability challenge to current architectures, since reliability requirements are less stringent than in a SAN (because if one server in the cluster fails, another can pick up the load), and management is more stringent than in a NAS.
Products: Panasas has designed its ActiveScale Storage System to meet these changing requirements, especially in the Linux world, with a 'storage cluster' design that it says offers the performance and reliability of SAN with the concurrency and manageability of NAS.
At its heart is the Panasas ActiveScale File System. It translates individual files into 'smart' data objects, which are then distributed across multiple storage hardware devices, called Panasas StorageBlades. Managing data as objects "which contain all application data, metadata and other extensible attributes" rather than the traditional approach of managing data as small blocks allows intelligence to be placed into the storage system, says Panasas. This includes things like dynamic load balancing, which can identify and migrate objects to less-utilized StorageBlades. The ActiveScale File System also stripes objects across multiple StorageBlades, offering either RAID 0 or RAID 5 protection for each file.
Additionally, its DirectFlow out-of-band data path technology provides direct communication between servers and StorageBlades following the installation of a client system driver.
These techniques allow parallel data paths to be established between Linux servers (or standard application servers or PCs) and StorageBlades, thereby avoiding the capacity and performance bottlenecks that exist with traditional monolithic file systems. Panasas claims this design provides it with up to 30 times the bandwidth of some alternatives (over 60Gbps), and with over seven times the random I/O performance of existing high-end NAS filers.
File system activity is managed using DirectorBlades. This is done out-of-band using DirectFlow, although in-band support is also available for UNIX NFS and Windows CIFS protocols (it does this by converting NFS and CIFS to DirectFlow, and claims record-breaking performance on these protocols, too). DirectorBlades "which hold up to 4GB of high-speed cache RAM and can be clustered" also virtualize data objects across StorageBlades, allowing them to be viewed as a single namespace, which Panasas says simplifies management. A PanActive Manager application provides volume management. Volumes can be created and assigned to specific applications, projects or users, and RAID protection levels can be assigned to individual volumes. It also provides reports on disk utilization, throughput and other metrics.
The upshot of this architecture is a system that provides near-linear scalability of performance with the addition of capacity. If more performance is required, Panasas says it's a simple case of adding more StorageBlades. Each blade is made from industry-standard components "including two Serial ATA disk drives (holding either 160GB [ATA only], 240GB or 500GB), cache memory, Intel processor and Ethernet NIC" and fits into a 4U shelf. Each shelf holds a mixture of up to 11 StorageBlades and DirectorBlades in any combination, with a total capacity of 5TB. It also comes with an optional integrated 16-port Ethernet switch blade.
Full redundancy with automatic failover and dynamic load balancing will be available early next year, along with support for Oracle clusters, 'nearline' storage capabilities and policy-based management. Further out, Panasas aims to take ActiveScale into the heart of the enterprise datacenter, with OLTP optimization, plus lifecycle data management and disaster recovery capabilities.
Business model: With such a radically different approach to storage, Panasas is selecting its target markets very carefully. It will start with the Linux computing space, and has tweaked its packaging for specific vertical markets, starting with the oil and gas, media, life sciences, and government science industries, as well as the emerging commercial Linux cluster market.
The company has secured seven paying customers to date, including Rockefeller University, the University of California at Berkeley, Sandia National Laboratories, NuTech and Geotrace, in addition to Los Alamos National Labs, which has already installed 120TB running ActiveScale for its simulation, visualization, management and planning applications, with an option to buy a further 500TB in 2004. Panasas says it competed with 17 other vendors, including all the tier one players, to win the contract. Aggressive pricing is also a key part of Panasas' value proposition, starting at $25,000 for 1.6TB, way lower than most comparable high-end NAS offerings.
It is currently building a small direct sales force, and is in the process of recruiting specialist VARs and integrators. OEMs are a potential route to market, although the company admits this is further out. More likely is that Panasas will form joint marketing and technology relationships with Linux specialists.
Competition: Although Panasas clearly will face competition from the specialist NAS vendors, not least of which are NetApp and EMC, in reality its approach puts it into conflict with just about every hardware vendor currently in the market. The company admits that its biggest challenge is to persuade customers to buy into its object-based design for storage, and it also admits that only 'pioneering' customers will be interested, at least in the shorter term.
No vendor can claim to have cracked the market for high-performance clustered NAS, although this is not for lack of trying. It looks like Panasas has learned from the mistakes that other failed ventures made. Scale Eight never made the transition from a service provider model, while both Tricord and Zambeel didn't offer support for CIFS until it was too late.
Meanwhile, plenty of other players are still targeting the clustered NAS market, albeit from slightly different angles. These include YottaYotta, Spinnaker Networks, Sistina, LeftHand Networks, Z-Force and ONstor.
SWOT analysis
Strengths Weaknesses
Panasas has a wealth of both technical and commercial management expertise, and has received a level of funding that is rare in the storage industry. With a significant customer deal already in the bag, it's off to a good start, too. Panasas' object-based design will confound all but the most progressive technical organizations. It has yet to secure any significant partnerships.
Opportunities Threats
With a strong initial focus on the Linux cluster space, Panasas can provide not just thought leadership but also product delivery in a significant emerging market. Many have tried to crack this market before, with little success. Panasas' radical architecture puts it into competition with all the major NAS and RAID vendors, all of which will seek to marginalize its proposition to ensure that it doesn't become mainstream.
