BOSTON — EMC Corp. this week at EMC World 2010 previewed data migration and storage efficiency features, as well as a common management interface for its block-storage Clariion and multiprotocol Celerra midrange disk array platforms.
EMC unveiled plans to offer the second version of its Fully Automated Storage Tiering (FAST) software with sub-LUN automated tiered storage support and block-level compression for Clariion and Celerra in July; they will also include support for flash as cache. The new Unisphere management console will be a big step toward consolidating the systems, although EMC executives deny they will end-of-life either of the existing platforms.
FAST 2 makes better use of flash
Clariion and Celerra are the first of EMC's disk arrays to announce support for FAST Version 2, which EMC said would be rolled out for all its disk arrays in the second half of this year. By adding support for tiering storage at the block rather than the LUN level, the new version makes for more efficient use of costly flash capacity and brings EMC up to speed with competitors such as Compellent Technologies Inc. and 3PAR Inc. FAST 2's support of flash as an extension of cache will provide rapid response to more bursty performance crunches that may arise in the array.
compression "squeezes out white space" from blocks, while data deduplication compares blocks against one another to find matches and throws out redundant data.
"Compression is much more simple but effective, and is particularly good for production storage," Sorenson said. "Data deduplication is probably better for large data sets like backup."
Industry experts are debating the ultimate value of these new efficiency features.
"All this time we storage people have been trying to be smarter than our storage — I love things like wide striping and automated performance tuning. In some cases they might do a worse job than a human, but in most cases, they'll do a better job," said Stephen Foskett, founder of Gestalt IT and director of Nirvanix consulting at cloud storage software vendor NIrvanix. Still, he added: "I wonder about the ultimate value of all these features — what if all you're doing is emptying space you've already paid for? Theoretically, you can use it to accommodate more workloads, but the reality is a lot of enterprises won't do that."
DeepStorage.net chief scientist Howard Marks countered that argument with, "Doesn't data expand to fill all available space?"
Unisphere: the first step in midrange disk array consolidation
Unisphere will first unite reporting and troubleshooting interfaces into one console, with sortable, customizable columns of system monitoring information. Subsequent releases of the product EMC previewed on the show floor will also add provisioning wizards in the idiom of applications (i.e., Exchange storage provisioning according to number of mailboxes rather than LUNs or volumes); direct connections to order and return replacements parts, as well as to EMC's online support and knowledge base from the management console; application-aware capacity reporting; a visual representation of the disk array backplane and drives with color-coded alerts corresponding to specific pieces of hardware within the machine; and automated snapshot scheduling and space provisioning.
Rumors about consolidation between Clariion and Celerra have been flying for months. During the company's most recent earnings call CEO Joe Tucci finally acknowledged consolidation plans, but EMC officials have also been emphatic that neither product will be discontinued.
Sorenson said there are plans to consolidate the arrays' replication tools. The two platforms could also unite on a common hardware platform with deeper integration than exists between the Celerra NAS head and Clariion block storage today. That led to speculation at the show about what form this might take. Possibilities floated at the show include Celerra's DART and Clariion's FLARE code running in separate cores of a multicore processor or in separate virtual machines (VMs) within one array, or consolidation of the separate code bases into one. Sorenson declined to say exactly how the consolidation might work, but said server virtualization and multicore processors are "good, new, enabling technologies for consolidation. For most customers, it's not 100% relevant."
One customer, however, said the details are important considering the number of changes planned for this update. "I hope they get it right the first time," said Eric Parson, senior systems support engineer at US Cellular in Schaumburg, Ill. "They've bundled a lot of stuff in here. Any one of those features could be a FLARE upgrade in itself. My initial concern from an operational perspective is how stable is it going to be out of the box?"
Parson predicted some customers will be uncomfortable with Unisphere. "It's going to be interesting to hear how they sell it," he said. "You have Clariion and Celerra people who are fairly set in their ways."
Parson said he's most interested in the new FAST cache option. "For something like the Oracle databases I've got, the indices would naturally migrate to that," he said. "Right now they're bigger than the CX4 cache." In the meantime, Parson said he's hoping to see that cache extend to 64 GB mirrored, twice the amount of mirrored cache currently supported.
Deepening VMware support
EMC is also rolling out support for VMware's vStorage APIs for Array Integration (VAAI), which offloads management of certain tasks from the VMware virtual server to the disk array for more efficiency and faster performance.
The VAAI integration results in four new array-based operations for VMware called thin provision stun, fast/full clones, write zero/write same and hardware offload locking. Thin provision stun pauses the virtual machine if thin-provisioned capacity on the disk array fills up. Fast/full clones offload snapshot or full block clone copying operations to the disk array rather than performing these operations in the VMware host. Write zero/write same lets the disk array fill up thick-provisioned volumes with zeroes. Hardware offload locking passes the SCSI reservation process (a bugaboo of VMware performance in very large environments) to the array.