EMC and Data Matrix Architecture. (Storage Networking).
It looks like that has changed. EMC has introduced the Symmetrix DMX, which it developed on top of an entirely new architecture: Direct Matrix Architecture. With this new product, which is fully compatible with existing Symmetrix models, operating systems and software, EMC has staked its claim again on the high-end storage market. Its timing is good: Even though the numbers of high-end storage customers may be shrinking due to budget concerns, the actual dollars that corporations are feeding into high-end storage remain very high. For example, high-end storage buyers are acquiring smaller companies that relied on mid-ranged arrays. Now that the larger company is bringing that data into the fold, they must protect and store the new data according to the same policies as the old, i.e., on high-end storage arrays.
In addition to offering very high performance, the DMX machines are fairly priced for their high-end market, starting at around $400,000 and ranging upwards of $2 million. (Yes, that's a lot of money, but high-end storage customers expect this kind of price tag. That's not to say they like it, but they will spend it.)
Competitors predictably downplayed the announcement while simultaneously announcing product enhancements of their own. EMC's two major competitors in the high-end storage market are HDS and IBM. (EMC happily reported a General Mills customer win over HP, but HP's high-end storage boxes are Hitachi boxes, not Storage Works.) HDS announced that it has doubled its Lightning 9900 V series capacity to 148TB of raw capacity, and is increasing connectivity up to 64 Fibre Channel 2GB connections. IBM made related announcements around its Shark array (ESS Enterprise Storage Server) high-end storage system. Shark now supports the Storage Management Initiative formerly known as CIM/Bluefin, and will expand within the month to a 72.8 GB disk drive capacity. In the face of these competing announcements, how does Direct Matrix Architecture differ from the earlier Symmetrix line?
Direct Matrix Architecture
Although DMX is integrated with the earlier Symmetrix models, it does not retain Symmetrix's bus architecture, nor is switch-based like the HDS arrays. EMC describes its new architecture as a matrix, a high-speed interconnect made up of dedicated point-to-point connections combined with global cache and an independent messaging and control function. According to Steve Kenniston, senior analyst for Enterprise Storage Group, DMX is a nicely timed and very significant new storage technology. "This is a good technology refresh for EMC. For 10 years they hadn't found a whole lot new, and now at the very high end they have a box that will outperform anything on the planet." Another analyst also commented that DMX will correct EMC's market losses due to flagging technology. Toni Sacconaghi at brokerage firm Sanford Bernstein reported in a research note, "Given that EMC is now competitively priced and has similar market share as Hitachi (and to some extent IBM), we believe its recent share losses are mainly attributa ble to its lagging hardware performance." DMX seems situated to answer that problem.
EMC released an upgraded version of the Enginuity operating system to handle the DMX hardware, but reports that only 5 percent of the code is changed. Users can also use existing Symmetrix software such as EMC's ControlCenter, and other software vendor packages supporting the Symmetrix will work on the DMX line.
Some sources insist on calling DMX a switched array, but its internal architecture is not a traditional backplane switching environment. Barry Burke, EMC's director of Symmetrix Platform Operations, talked about the confusion. "In general, the requirement of a storage array is to switch, so in a way every storage array is one big giant switch. What we're talking about is the board that talks to the hosts and disk -- you can use a bus, you can use a switch or a bunch of switches. What we've done is eliminate those and replaced them with 128 private highways."
The matrix contains up to 128 point-to-point connections, each of which directly link each front-end channel director and back-end disk director to all global cache memory regions. This provides the channel and disk directors with high-speed, non-blocking parallel access to all global cache directors simultaneously, which nearly eliminates contention, latency and bandwidth constraints. Here are the numbers: EMC reports that each dedicated connection is capable of transporting data at 500 MB per second, resulting in a total data path bandwidth of 64 GB per second.
The global cache director is a multifunctional, solid-state global memory technology that manages data movement between all channel and disk directors. The global cache directors optimize DMX to handle unpredictable activity bursts as well as sustained performance and throughput. Symmetrix DMX systems can have a maximum of eight global cache directors providing shared global memory. Each cache director contains four independent regions of cache memory for a total of 32 separately addressable, simultaneously accessible regions of cache memory. EMC reports that each region can deliver 500 MB per second of throughput, enabling 16 GB per second throughput for the entire global cache memory. The cache directors combine data offload engines, dynamic cache optimization technology and memory striping techniques.
These cache directors also provide fault tolerance and high availability through internal component redundancy, end-to-end integrity validation and advanced error mitigation. The error mitigation tool, tongue-trippingly called the Triple Module Redundancy with Majority Voting, isolates components by eliminating data and messaging busses and switches, and provides continuous error detection, fault isolation and phone-home diagnostics. DMX also removes the control function from the data transport matrix interconnect contains a separate, standalone matrix interconnect to handle component communication and synchronization. EMC is positioning DMX as a next-generation architecture, while still a fully compatible Symmetrix array. EMC plans to build global cache capacity and regions using DMX, while supporting 10GB Fibre Channel disk drives or 2 GB processors.
* Symmetrix DMX800, a rack-mount high-end storage system, scales from 8 to 16 front-end ports, from 1.2 to 17.5 terabytes of raw capacity and from 4 to 32 gigabytes of global cache for open systems environments.
* Symmetrix DMX1000, a single-bay integrated system, scales from 8 to 48 front-end ports, from 3.5 to 21 terabytes of raw capacity (3 to 18.5 usable) and from 4 to 64 gigabytes of global cache for main-frame and open systems environments.
* Symmetrix DMX2000, a dual-bay integrated system, scales from 8 to 96 front-end ports, from 7 to 42 terabytes raw capacity (6.1 to 37 usable) and from 8 to 128 gigabytes of global cache for main-frame and open systems environments.
The DMX800's rack mount capability is especially interesting, since it avoids the extra cost and labor of building or expanding raised-floor data-centers. It also offers the high-end market the advantages of modular architecture and distributed functions of the mid-ranged arrays. EMC's CLARIiiON line, for example, provides distributed services to the network then feeds the data to the back-end Symmetrix arrays. In high-end environments, the DMX800 can provide the distributed services as a fully integrated Symmetrix. And since EMC used the same disk array enclosure for both the DMX800 and the CLARiiON CX600, users can convert existing CLARiiONs to DMX by swapping controllers.
Reasonable high-end pricing is a part of EMC's strategy. They want to stop making deep customer discounts, so by pricing the new Symmetrix more affordably they have a better foothold in customer negotiations. The Symmetrix DMX comes in at a relatively economical 4 to 8 cents per megabyte ($40 to $80 per GB). Since the Symmetrix packs a big megabyte punch, that works out to $409,000 to $2.5 million list price per Symmetrix DMX.
According to Sageza analyst Charles King, data storage is not so much about sort but about access, that is, how easily and effectively storage underwrites business processes. If Symmetrix DMX can give EMC a stronger footing in this area, as well as remain integrated with previous Symmetrix models, it might regain lost market share from HDS and IBM. The next year should prove interesting as those two competitors keep sniping at DMX and raising the bars on their own high-end storage systems, meanwhile battling it out in the well-heeled mid-tier market.
EMC concludes that capacity, performance and connectivity gains must provide the service levels customers need. Burke said, "When and if you see EMC come out to meet a higher capacity model, a key part of that will be maintaining high service levels for our high-end customers. Taking snapshots, replicating data, protecting data, reallocating drive spindles -- these things take up CPU and bandwidth power. A customer wants that service level and capacity, and this is the challenge for this new marketplace. Get ahead and stay ahead of the processing power of the network."
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||EMC Symmetrix DMX|
|Author:||Chudnow, Christine Taylor|
|Publication:||Computer Technology Review|
|Date:||Feb 1, 2003|
|Previous Article:||Active archiving and HSM: best-practices. (Storage Networking).|
|Next Article:||Western digital's raptor enters the enterprise storage market. (Storage Networking).|