Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
This is a different buyer’s guide than our usual fare. Normally, we cover existing products. This time, however, we are showcasing a brand new EMC offering as well as data on a couple of up-and-coming products courtesy of a demo and advanced look provided at the recent VMworld.
The common denominator is flash. EMC is going all out with its flash portfolio and will soon have a lineup that ranges from hybrid arrays (mixed SATA and Solid State Drive –SSD) to all flash arrays and server-based PCIe flash.
“EMC’s strategy is flash everywhere,” said Barry Ader, Senior Director of Product Management for flash products and technologies.
EMC released VFCache in the first quarter as server flash caching to reduce application latency and increase throughput. It is a combo of caching software and PCIe flash that puts the data closer to the application. It caches the most frequently used data on the server-based PCIe card, which proves particularly handy for read-intensive applications.
As VFCache sits in the server on the PCIe bus, the data doesn’t have to travel through the network to the storage array so response time is fast.
This latest 1.5 version announced today adds deduplication technology to VFCache as well as more capacity and form factors. The deduplication technology comes from EMC. With Microsoft Exchange, for example, Ader said a 360 GB of flash could be reduced by 20% or more – requiring 300 GB.
“Dedupe also means you are writing to the card less often so its life is extended,” said Ader.
Interoperability with VMware vMotion has been added to VFCache to enable ease of migration and VM maintenance. For a recovery, for example, system downtime is no longer required.
Card capacities up to 700 GB have been added along with the ability to have multiple cards per server and improved tuning of caching algorithms. Support has been added for active/passive clusters, more OSes (even for Microsoft Hype-V), more servers and more protocols.
And there is now support for Cisco UCS B Series blade servers in combo with LSI Nytro WarpDrive PCIe mezzanine cards. In this later case, 400 GB and 800 GB flash can be ordered through Cisco in the fourth quarter of this year.
On the management side, VFCache provides EMC VMax storage administrators with more operational information, such as VFCache’s relationships to specific LUNS, more performance statistics, and error condition reporting.
What else lies ahead?
“For 2013 we will introduce more integration with EMC storage,” said Ader. “This includes unified management for VFCache and EMC VNX.”
Since its acquisition of XtremeIO a couple of months ago, EMC has released few details. But at VMworld, the company unveiled a live demo and showed off the box.
This comes under the banner of Project X, which is an all flash array, i.e. it has only flash inside and no disks. During the demo, the company showed almost 1 million reads and about 500,000 writes.
Project X is based upon what EMC calls “X-Brick” building blocks (see photo which shows four X-Bricks at the top), which uses off-the-shelf hardware comprised of standard Intel x86 components. As you add another brick, the number of IOPS increases proportionately.
Up to eight bricks can be joined together into a single system with InfiniBand used for internal communication. Otherwise Fibre Channel (FC) or iSCSI is utilized to communicate to the host.
The secret sauce is the XtremeIO software, which organizes data into 4k blocks and maintains latency under 1 ms.
“The system can recognize if a 4k block has been seen before and won’t rewrite it,” said Josh Goldstein, VP marketing and product management at EMC. “It is essentially a content-based engine that is inline, global and always on.”
It also includes a flash-specific data protection scheme that Goldstein said offered 1.6 times better performance than RAID 1. The user can lose two drives and not lose data, while only consuming 8 percent of the available capacity for data protection. The system also requires no fragmentation or garbage collection, and is fully integrated with VMware.
Another feature is that it can drastically reduce the storage requirements for databases and VMs. Say, for example, there are thousands of VMs running. Many will have similar or the same configuration. Instead of storing a copy of each, the XtremeIO software tags the metadata and only needs to record one copy.
“One customer went from hundreds of TB for databases down to 17 TB, as there is no longer any need to store individual clones and snapshots that take up loads of space,” said Goldstein.
Project Thunder and the Rest
Project Thunder is a little more mysterious. EMC’s briefing didn’t really give too much away. It says Project Thunder is a purpose-built, low-latency, server networked flash-based appliance that is scalable, serviceable, and shareable.
It is apparently supposed to deliver I/Os measured in millions and timed in microseconds. This flash appliance is said to be optimized for high-frequency, low-latency read/write workloads. However, it will operate at the server level and not within a storage network. Stay tuned for more details.
“Project Thunder will be out in 2013 but is already in some customer environments,” said Ader.
Finally, EMC delivers what it calls hybrid arrays for its VNX, Isilon and VMax product lines. These make use of a combo of SATA and SSD. A tiering set up is arranged by EMC Fully Automated Storage Tiering (FAST) software to determine which data is on flash and which is on disk.
“FAST made flash take off as you could leverage the cost per GB of SATA with high performance flash to get better value per IOP,” said Ader.