Dell has built a tremendous reputation around primary enterprise database and virtualization environments thanks to the capabilities of EqualLogic and Compellent. But it’s hard to become the leading storage OEM without strong file capabilities, which are needed by those same enterprise customers for user-shares and other workloads, as well as for those Petabyte-scale customers in specific markets like Media, Healthcare, and Energy. These guys are already buying truckloads of PowerEdge servers and have been asking Dell to deliver a full featured NAS capability with great value…perfectly in-line with the Dell brand.
Three years ago we began investing in NAS and object storage, acquiring the assets and dev team from scale-out storage innovator Exanet after technical due diligence on competing candidates. We extended that code into what’s now our Fluid File System (FluidFS) that forms the NAS backbone across our entire storage portfolio. We delivered our 1.0 products a year ago, and we recently announced 3 new models coming later this Summer, all tightly integrated with our Dell product families; The Compellent FS8600, the EqualLogic FS7600, and the PowerVault NX3600. These products will still provide the flexibility of integrating into your SAN as a separate physical gateway, and for ease-of-use our goal is to deliver a fully unified management experience. These new products will include a purpose-built NAS appliance that provides improved reliability and performance, connectivity for 1G, 10G, iSCSI, and Fibrechannel, and an internal cluster interconnect. But the real value of the new NAS products is in FluidFS itself, which is truly a file-system for future generations.
No Namespace or Scalability Limits
Architecturally there are no practical limits on FluidFS scalability. This means the capacity limits of our products are really only a function of the underlying SAN capacity, or what we chose to test to meet the strategies of a given product. The best piece is no restrictions on name-space sizing. Today products like EMCs VNX have strict limits on capacity (256TB per NAS controller), volume size (16TB) and file size. Even with Netapp’s new clustering capabilities, Ontap has similar volume scalability limitations. On the other hand, FluidFS today can put the entire capacity of the underlying SAN in a single pool, a single volume, and even in a single directory (if that’s what you really want to do). We realize not everyone has a Petabyte of files, but without name-space restrictions, Sysadmin have just one less thing to work around.
Scale-out clustered architecture allows scaling on-demand
A core tenet of Dell’s Fluid Data Architecture is peer scaling; the ability to grow easily and transparently as the customers’ workload demands increase. FluidFS provides performance scaling by adding controllers to the cluster, and customers can grow capacity by adding SAN arrays. The advanced clustering model of FluidFS allows for namespace ownership to be allocated across cluster members without the need for central metadata or “housekeeping” server. But these mechanics are transparent to the clients because FluidFS provides a single virtual IP address (VIP) and provides native load balancing for distribution of client requests across the cluster. To-date we’ve certified up to 8 controllers in a cluster, which when outfitted with 10G LAN uplinks can stream data at near-HPC rates.
High Performance to meet the needs of all workloads
Some legacy users of the Exanet ExaStore system have told us that this file system is the fastest NAS solution they have ever tested, even above some HPC solutions like Lustre. Even though it’s not our immediate goal to go put 1000 SSDs in a cluster and shoot for 1 million SpecSFS IOPS, that strength was inherited by FluidFS. In addition to the ability to add NAS controllers on-demand, we can point out a few key features delivering performance:
1. FluidFS employs a cache centric design. As soon as new writes land in cache, they are protected through pair-wise mirroring and acknowledged back to the client. This means that write-intensive workloads can happen at wire speed, such as capturing data-streams from a next-generation DNA sequencer. For read-intensive workloads like video streaming, FluidFS again delivers wire-speed performance by distributing hot objects across all the caches of all the IO controllers as well as leveraging underlying SAN optimizations such as Compellent Data Progression.
2. Works great even for small file workloads: Traditional NAS systems often faced a tradeoff between namespace scalability and small-file capacity efficiency (both capacity efficiency and performance overhead. For example the EMC/Isilon product is optimized for large-file workloads. FluidFS handles both large and small file workloads with no sweat. A couple of the features that give us this performance are write coalescing (holding off on data write-back until we have 1MB of data in queue), the storing of very small files in inodes, striping of large files across LUNs, and redirect-on-write snapshots that greatly reduce file-system fragmentation over time.
Feature Rich Capabilities
What really made it clear to Dell that the Exanet code-base was the right starting point for us, was that unlike other clustered high performance NAS solutions, all of the hard problems had been solved, such as reliability of snapshots and replication in a clustered environment under degraded conditions. HPC-oriented solutions – for example those relying on Samba for windows support – failed to meet the security and authentication requirements of enterprise customers. FluidFS now provides the best of both worlds. Some of the key capabilities supported by FluidFS today include native ACL/SID support, cross-protocol security mapping, efficient snapshots including end-user restore, diff-based replication, quotas, NDMP, and anti-virus support.
Future Directions for FluidFS
Dell is continuing to invest aggressively in FluidFS as our core IP for unstructured data workloads. We’re being careful to deliver the most important new features first, and we’re regularly checking with customers to make sure we have the right priorities in mind. It’s not interesting (or strategically wise) to merely copy all of our competitor’s features, but to also look for ways to solve new customer problems in a creative way, and deliver on the promise of the Fluid Data Architecture. So while we are committed to delivering important capabilities like next generation protocols and multi-Petabyte scalability, we’re also planning on things like richer tiering capabilities, advanced compression, and better-together strategies with other Dell products like the DR4000 and Dell’s cloud services.
