EMC Corp. today rolled out a range of enhancements to its midrange Clariion and Celerra storage array platforms to drive common management, more efficient use of solid-state drives (SSDs) and primary data reduction. These include features it previewed earlier this year such as sub-LUN-level automated data tiering, block compression, a unified management application for Clariion and Celerra, integration with VMware's vCenter and support...
for VMware's vStorage APIs for Array Integration (VAAI).
EMC also rolled out two new Celerra NAS gateways, and said it would lower pricing on SSDs and offer native Fibre Channel over Ethernet (FCoE) for Clariion and Celerra. The NAS gateway, new SSD pricing and FCoE support had not been previously announced. EMC previewed the Unisphere management console, an upgrade to its Fully Automated Storage Tiering (FAST) software with a FAST Cache option for SSDs, block compression and VAAI support at EMC World 2010 in May.
Unisphere management software for Clariion and Celerra replaces Navisphere for Clariion and Celerra Manager. Common management is the first step toward consolidating the Clariion SAN and Celerra unified storage platforms, although EMC continues to sell the storage platforms separately.
Clariion customer Joshua Lukes, a Raleigh, N.C.-based IT support manager at energy firm Itron Inc., tested Unisphere and said he was impressed by its GUI and simplicity compared to Navisphere.
"This finally brought EMC out of the '80s," Lukes said. "I've been using Navisphere since it was a Data General product, and it looks almost identical today. Unisphere is much faster and more intelligently laid out. The dashboards answer questions I typically have, and things I need to go through 20 clicks to figure out [with Navisphere] are right on the front page. It definitely speeds up provisioning of storage."
Brian Garrett, vice president, Enterprise Strategy Group (ESG) Lab, was part of an ESG team that tested EMC's new unified storage capabilities, and was also impressed by the new Unisphere.
"The look and feel of Unisphere was a pleasant surprise," Garrett said. "This looks and feels like a fresh approach. It's a Version 1 GUI, but the workflow, wizards, and look and feel was a lot more mature than I expected."
Itron's Lukes said EMC still has work to do with Unisphere's monitoring, however. "There needs to be more built-in monitoring and better performance analysis," he said. "[Navisphere] Analyzer is an add-on product, and it's still back in the '80s. It gathers tons of data but it's not easy to get that data out of the system. I want to know things like if I'm exceeding the capacity of this LUN and need to add some disk and migrate data off of it."
EMC FAST and FAST Cache
EMC is finally offering sub-LUN automated tiering in the EMC FAST software it first launched last December. While it has informally referred to the upgrade as FAST 2 while previewing its capabilities over the past year, the shipping version of the product is still known as FAST.
FAST re-balances the storage pool daily to make sure hot data is moved to solid-state drives, and cold data goes to SATA or other lower cost storage. The FAST Cache option supports flash as an extension of cache, and makes sure the hottest data is stored on SSDs to improve performance.
EMC customers have been waiting for sub-LUN tiering to take full advantage of SSDs, while competitors -- mostly notably 3PAR, Compellent and IBM -- beat it to the punch with the more granular tiering capability.
Itron's Lukes said Fast Cache is the new feature he's most interested in. "We got very positive results," he said. "We had a 10-time performance increase on some of our LUNs where we can throw FAST Cache at them. We easily exhausted our regular cache in Clariion, and Fast Cache picks up the slack."
He said his RAID array without Fast Cache normally works at approximately 500 IOPS but reached 5,000 IOPS with FAST Cache, although that was with small data sets. "I suspect that improvement would erode if we threw a 20 TB database behind it where numerous LUNs are hitting that cache and filling it up, but it's still a nice improvement," he said.
As for EMC FAST, Lukes said it would be useful for some data he has. "We have meter reads that are very important for the first 35 days while we're doing monthly readings and then it gets less and less important, but we still need to keep them for regulatory reasons," he said. "FAST is useful for starting that off on flash and moving it down slowly as it ages out."
New solid-state drives
EMC is also rolling out new 100 GB and 200 GB solid-state drives for Clariion and Celerra that it claims cost up to 30% less than previous drives. Jon Siegal, EMC's director of product marketing, said he expects the new pricing and FAST capabilities to drive greater SSD adoption.
"A lot of customers wanted to take advantage of SSDs but haven't had the means," Siegal said. "We're helping to take flash to the masses in the midrange market."
Itron's Lukes said his Clariion CX4-960 has six 72 GB drives, and EMC loaned him a system with 400 GB drives for testing. He said the performance was good, but he's not sure he could afford that much flash even with the price drop. The cost of the 10 400 GB drives he tested would cost approximately $350,000, he said.
"Performance is attractive," Lukes said, "but the cost of the flash drives is the unattractive part."
EMC has offered file compression on Clariion and Celerra for more than a year, but now adds block-level primary compression. EMC's compression lets customers compress or decompress LUNs with one click to reduce capacity of less-frequently accessed LUNs. EMC claims a 50% average reduction, with up to a 77% reduction on text files.
The primary data reduction market is heating up. Although NetApp has offered primary deduplication for several years, other vendors are lining up their products. Dell Inc. acquired dedupe vendor Ocarina Networks, IBM bought compression vendor Storwize last month and Hewlett-Packard Co. said it would expand its StoreOnce deduplication feature to primary storage. EMC is the backup deduplication leader with its Data Domain and Avamar products, but chose to go with compression for primary data.
Greg Schulz, founder and senior analyst at StorageIO Group, said EMC made a smart decision to go with compression instead of deduplication for primary data. Deduplication has a higher reduction ratio than compression, he said, but it also negatively impacts performance. He pointed out that a 100 TB backup store that gets a 2-1 compression ratio still nets a significant storage reduction.
"The real diamond in the rough of this EMC announcement is compression," Schulz said. "Everybody's talking about primary dedupe, the industry's drunk on dedupe -- for any malady, dedupe it. But data deduplication introduces latency, time delays. For primary data, it's about performance. Compression is the home run for the primary space. It's more about performance and data movement rates than reduction ratios."
ESG Labs' Garrett said the benefits of compression vs. deduplication depend on what type of data you're handling. "Databases and email are compressible because of repeatable patterns," he said. "But some data's already compressed, like video or audio files so compression doesn't help as much."
Itron's Lukes said he hadn't done much testing of the block compression because Itron uses its CX4-960 mostly as a repository for Oracle and Microsoft SQL databases and "compression doesn't fit into that scheme too well," he noted.
NAS gateway, Fibre Channel over Ethernet
EMC launched two new NAS gateways, the Celerra VG2 and VG8, which use the new 64-bit Intel Xeon 5600 processors, code-named "Westmere." The VG2 and VG8 will replace the Celerra NS-G2 and NS-G8 gateways.
EMC will also offer an add-on module providing native Fibre Channel over Ethernet support to its Clariion and Celerra arrays. NetApp is the only storage vendor with native FCoE support on its storage arrays.
Although the storage FCoE ecosystem is still being built out, Lukes said he's interested in FCoE because "the management of our Fibre network has become not very nice. We use blade servers and each blade chassis has to have two Fibre Channel switches and connect into one of four core switches, and it's just become a cascaded nasty mess. We have hundreds of zones and it's difficult to manage. If I need to make a change in one machine, it affects other machines. I need to get it to where it's like adding an IP address."