I’m a closet storage geek. I don’t have a lot of clients that need really cool storage products, but I wish I did. When I can, I try to read up on as much information about the storage world as possible. I’ve come across three fascinating players in the flash storage world that any SysAdmin should keep their eye on.
Pure Storage has a rather daring claim. Their arrays are claimed to be ten times faster, smaller and more power efficient than disks.
The product touts its software as much as (if not more than) its hardware. The software is called the Purity Operating Environment and performs global deduplication, compression and thin provisioning to make data storage more efficient. From their site:
Purity is a fully-virtualized storage operating environment, which abstracts individual flash devices into a single unified storage pool and optimizes data placement across the pool. Purity’s data layout is aligned with the erase block size of the flash, reducing flash write amplification to extend flash life and improve performance. Moreover, Purity’s data structures are “append only”, meaning that all writes (new data, updates, parity re-builds, recoveries) are coalesced into write segments that are always placed somewhere new, improving performance and extending flash life. Finally, Purity implements a set of active background flash management services (wear leveling, deletion management, performance optimization, integrity/health checking and automatic healing) across the global pool to ensure the reliability of both the data and the underlying flash.
As of this post they only offer two appliances. A 2 controller / 2 storage shelf version and a 1 controller / 1 storage shelf version. Pictured below is the single controller / shelf model:
The Pure Storage FlashArray is built on a flexible, scalable, redundant, highly-available hardware architecture, designed to allow Pure Storage solutions to scale from single application to consolidated cross-data center deployments. The Pure Storage FlashArray implements a node-based design, with clustered controllers and storage shelves. This allows for the independent scaling of storage performance (controllers) and storage capacity (storage shelves). Configurations can range from 10s to 100s of usable TBs of flash storage, for both HA and non-HA configurations.
The Pure Storage CEO is none other than Scott Dietzen former president and CEO of Zimbra. The CIO is John Colgrove who worked for Veritas. Also on board is Michael Cornwell who worked on flash chips for the iPod and iPhone while at Apple. He has also worked on flash-based products at Sun.
Pure has attracted some important investors including Mendel Rosenblum and Diane Greene, the couple who founded VMware. Also funding Pure Storage is Greylock Partners which brings Pure Storage’s total amount of VC funding to about 50 million dollars.
I foresee big things for Pure Storage and hope to work with of their equipment some day.
Fusion-io doesn’t actually make server appliances. They make flash devices that interface with a computer through the PCI bus. They make three major hardware products, the ioDrive, ioDrive Duo and ioDrive Octal (the latter shown below).
The Fusion ioDrive can supply 160GB (at 123,000 Mixed IOPS [75/25/r/w]) to 640GB (at 74,000 Mixed IOPS [75/25/r/w]). The ioDrive Duo can supply 320GB (at 238,000 Mixed IOPS [75/25/r/w]) to 1.28TB (at 150,000 Mixed IOPS [75/25/r/w]). Finally, the ioDrive Octal can supply 5.12TB at 729,000 75/25 Mixed IOPS (512 B).
Fusion-io went public in the summer of 2011. They have been invested into by Samsung and are working with the major flash chip manufacturer Toshiba.
They make a software platform called ioSphere that allows you the following features (taken from their website):
- Manage all ioMemory modules across all servers from a single interface
- Historical performance monitoring
- Warranty forecasting
Also available is the software tool known as direct cache. From the Fusion-io website:
Fusion’s directCache transforms ioMemory into a transparent, auto-tiering, acceleration device to cache any block-based storage medium whether it is a disk array, SAN, direct attached storage or iSCSI target. directCache places the caching software in the server to deliver lower cache latency. With directCache, Fusion-io customers can have Terabytes of cache acceleration at their fingertips to speed performance of any backing store. This add-on module for ioSphere integrates tightly with Fusion’s Virtual Storage Layer, a flash-optimized OS subsystem, to deliver immediate application workload performance improvements.
They look like a great tool for anyone doing CAD work or video editing. I’m not sure about building out a server full of them though. Perhaps I’m spoiled by the pretty boxxen that most vendors will supply.
The former CEO of Fusion-io was Donald Basile. Why is that important to Violin Memory? Because he became the former Fusion-io CEO when he left to head up Violin Memory. Unlike Fusion-io, Violin Memory is focused on datacenter products. With an impressive catalog of appliances, I won’t go into depth on each of them here, but will give a brief overview (using some or all of the text from their product website):
3200 Flash Memory Array – A redundant, modular 3U memory array that scales from 500GB to 10TB SLC NAND Flash
vCACHE NFS Caching – The NFS caching system is built on flash memory arrays in conjunction with their vCACHE software. These vCACHE NFS Caching systems increase the size of the caching available to applications, but also reduce the cost per GB of the cache by more than 70%. Unlike an internal cache, the vCACHE system can also support 200K operations per second. vCACHE systems enable the entire active data set to be stored in cache. This may be 5%, 10% or 20% of the total data stored in the filers. By caching the entire data set, the full application speed-up of 5x to 30x is enabled for both IOPS and latency. With caches of 1% or less, the speed-up is typically a small increase over the standard disk speed.
3140 Capacity Flash Memory Array – A redundant, modular 3U memory array that scales to 40TB of Capacity Flash. It scales to more than 500TB in a rack with performance over 1.5 Million IOPS. The Violin 3140 includes hardware-based flash RAID across hot-swappable memory modules to provide data protection and high-sustained IOPS.
SAN Attached – The Violin Memory Arrays can be clustered via PCIe with one or more Memory Gateways that provide connections via an combination of Fibre Channel (8Gb/s or 4Gb/s), 10 GbE (iSCSI or FCoE), or InfiniBand. A single Memory Gateway can support up to 4 Violin Memory Arrays, 400K IOPS and over 3GB/s. Through striping, each LUN on the system can get the full bandwidth and IOPS capability of the cluster. LUNs can range in size from 1GB to 120TB!
DRAM Array – Mmemory appliances made to provide a platform for provisioning DRAM as a large scale Tier-Ø storage infrastructure.
Violin provides “vCLUSTER Management” which is their storage management software to keep an eye on their products in your environment.
Flash storage is going to inevitably replace spinning disks, and it appears that with the above offerings, especially Pure Storage, it’s going to happen in this decade. Do you have any experience with the above systems? Or perhaps another upstart flash storage vendor? Let me know in the comments below.