Thursday November 14 EMC made its long awaited announcement of general availability of the first version of the XtremIO all flash array technology, 18 months after it purchased the flash storage startup and nearly eight months after starting a limited release to a few favored customers. Overall the initial announcement is less than overwhelming, considering it comes from the dominant storage provider, a company focused solely on storage. Both IBM and Hewlett-Packard, which have had all-flash arrays with greater capacities and competitive performance on the market for several months, issued reactions before the announcement.
The initial version, which will ship November 19, is based on a 10 TB “X-Brick” that can be sacked into clusters of up to four bricks or 40 TB total raw storage, says Ehud Rokach, vice president and general manager of XtremeIO. It ships with thin provisioning and fine grained in-line deduplication, but no data compression.
Rokach says the deduplication can increase total storage by 5X to 50X depending on the use case by only storing one copy of each 4 kilobyte block of data on the array. EMC promises that this deduplication does not add significant computing overhead to the system. It also comes with data security and with EMC’s VPLEX storage control and data tiering built in. Rokach promises that an upgraded system, based on a 20 TB X-Brick with a total capacity of 80 TB per cluster will be introduced in 1Q14, and that ViPR support will be added in 2014 as well.
The array also features content-based data placement, a dual-stage metadata engine, XtremIO Data Protection (XDP) that guards against SSD failures and replaces RAID in the system, and shared in-memory metadata. EMC claims that XDP delivers up to 4X better performance and flash endurance than RAID and allows customers to use 100 percent of the array capacity.
VPLEX, Rokach said, allows the array to act as Tier 1 in an automatic multitiered environment with other EMC arrays. VPLEX can demote inactive data to less expensive disk arrays based on custom business rules, leaving the array free for new data or new uses.
Comparing Systems
On its face, the announcement is disappointing. For instance, the IBM FlashSystem 820 tops out at 1 PByte of raw capacity, making XtremeIO’s 40 TB, or even the promised 80TB, per array puny by comparison. IBM also argues in a reaction statement that it uses eMLC NAND flash directly on the system motherboard, providing 10x higher transfer speeds than the SAS interface that XtremIO uses to connect the motherboard with its SSD Flash. The result, according to IBM, is that its system has 110us latency while XtremIO has 500us latency. The IBM FlashSystem can also support 500K IOPS compared to XtremIO’s 250K IOPs. The IBM array uses 400W of power versus 750W for XtremIO.
Rokach replies that the important thing is the transfer rate of the array as a whole and said “I am confident that XtremIO will stand up to the performance of any of its competitors.”
HP Storage VP of Worldwide Marketing Craig Nunes gave a detailed criticism of the announcement in a statement released soon after the announcement. Among his criticisms:
Insufficient product density: XtremIO provides only 1.7 TB of capacity per rack unit compared to 4.8 TB per rack unit for the HP 3PAR StoreServ 7450 all-flash system. This, he says, makes the new XtremIO system “problematic in today’s space- and power-constrained datacenters.”
Limited scalability: The initial system only scales to 40 TB. “Sales of the flash-optimized HP 3PAR 7450 Storage demonstrate customer demand for all-flash systems with raw capacity that is more than 2x the maximum capacity of EMC’s new array.”
Silo Creep: XtremIO adds a sixth storage architecture to EMC’s lineup of VMAX, VPLEX, VNX, VNXe, and RecoverPoint. While these can be unified at the management level with VPLEX, the underlying architectures still have to be managed separately using different commands and skill sets.
He also says that XtremIO appears to lack data replication capabilities, required for mission-critical enterprise environments, and advanced data services such as Quality of Service controls needed for IT-as-a-Service in heavily virtualized ecosystems.
Use Cases
Rokach says that the eight-month limited trial was a complete success that proved that the XtremIO is a strong choice for production environments. He provided several use cases for the array:
VDI: Rokach said that XtremIO is seeing “a strong ramp-up of demand for VDI”, confirming similar reports from other service providers and vendors. Flash is absolutely necessary for VDI to eliminate the long delays caused by the morning sign-on and evening sign-off storms that overwhelm disk systems with their limited IO. Rokach says XtremIO is particularly well suited for VDI in part because of its in-line deduplication. In production environments it achieved 50:1 or more data dedup, giving it a more attractive price-point than competitors without in-line dedup. He also cited strong administrative support. “You can provision a thousand more desktops on a single EMC XtremeIO X-Brick while at the same time maintaining 2,000 existing users on the same X-Brick.”
Virtualized server environments: Large virtualized environments have substantial IO challenges and in particular a lot of random IO. Flash arrays, with their huge IO advantage and order-of-magnitude speed advantage handling random IO, are ideal for supporting these environments.
Analytics and OLTP: Customers need faster response times for analytics as the pace of business continues to increase. Some EMC customers are realizing important performance increases running large Oracle and SAP databases on XtremIO arrays.
Test/dev: One early XtremIO user was able to combine production and test/dev for a major database on the same XtremIO brick while increasing the speed of the production system. This company actually could increase the number of database copies supporting test/dev, making its development process more effective while increasing speed of the production system, thanks largely to the in-line data dedup which meant the extra copies were essentially supported for free.
Conclusions
The XtremIO system is definitely a good solution for some applications, as demonstrated by the eight-months of production trials by EMC customers. At the same time it is not an industry leader, and this is not the announcement that the industry looked for from the market leader. Is it enough to protect EMC’s dominance in the storage market? Maybe, but EMC has definitely left the door open here, and both IBM and HP are committed to increasing their presence in the storage market, largely at the expense of EMC.
IBM is a particularly dangerous shark in this bloody water. Its resources dwarf EMC. For decades it paid little attention to storage, basically selling storage “after it had sold everything else” in the words of Dr. Ambuj Goyal, general manager of IBM System Storage and Networking. That attitude changed in the last two years, and in particular with Dr. Goyal’s appointment to head IBM storage. HP is also focused and determined and provides strong technology.
EMC has to stop worrying about protecting its existing empire of disk storage and realize that it is in real danger of becoming a legacy vendor in the new reality of flash. As the price of flash drops and its advantages become compelling for increasing numbers of business use cases, it needs to be fully competitive with IBM, HP, and the all-flash array vendors to hold its position.
For EMC customers the message is certainly put XtremIO on your short list for next-generation storage. But also consider other vendors’ arrays and do not hesitate to play them against each other. This is a buyers’ market, and strong negotiation can get users the right system at the best