The rise of solid-state drives (SSDs) in enterprise data storage arrays last year brought about an emphasis on...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
automated tiered data storage software to better take advantage of expensive SSD capacity. Compellent Technologies Inc. has had its Data Progression software out long before it started shipping SSDs, but competitors including EMC Corp. and Avere Inc. now say they've built better mousetraps.
Compellent: Batch jobs according to user policy most performance-efficient
Compellent's Storage Center SAN automatically tracks the use of blocks of data on the system and moves them between tiers of mechanical disk, SSD and RAID protection level. Customers set policies and schedule movement of data during low activity periods in the data center.
EMC Corp. last December came out with its Fully Automated Storage Tiering (FAST) software primarily for its high-end Symmetrix V-MAX arrays. EMC claims FAST can migrate data between tiers in real time. Compellent director of product marketing Bob Fine said he's skeptical about whether real-time migration will live up to the hype.
"The idea of real-time data movement is great because of the promise of speed and efficiency, but from our viewpoint, there are a lot of key tradeoffs," Fine said. "Today's drives are still mechanical, so data movement is a lengthy process of reading from one drive and copying to another. Some solutions that promise real-time data movement also make copies in multiple locations, which takes up more disk space and eliminates some of the most significant benefits of automated tiered storage: disk utilization and efficiency."
Fine estimated that given a 50 MBps transfer rate, "moving one terabyte of data would take about five-and-a-half hours to read data from one disk drive and simultaneously write the data to a new drive."
EMC claims the horsepower to move in real time
An EMC spokesperson responded for request for comment on the disk transfer rate bottleneck in an email to SearchStorage.com, and said EMC can also run FAST data movement as a batch job if the customer prefers.
But the spokesperson dismissed Fine's point about the migration time for a 1 TB LUN. "You typically don't put a terabyte LUN of data on a drive, as a LUN normally exists on a number of drives," the EMC spokesperson countered. "So a LUN [logical unit number] movement involves multiple drives times some number of megabytes per second per drive. With just eight drives, you could easily move nearly 1 TB per minute, not the hours they suggest.
"V-Max can do this LUN movement for many LUNs spanning many drives concurrently. You need lots of processors and bandwidth to scale like that. In Compellent's case, they only have a small amount of processors and bandwidth, and limited SW intelligence, which requires batch jobs."
Avere: a hybrid approach
Newcomer Avere also uses parallelization among disk drive groups and nodes in its scale-out network-attached storage (NAS) architecture to provide real-time migration between the top tiers of storage in its system.
Avere director of product marketing Jeff Tabor said Avere's FXT storage systems use solid-state tiers, DRAM, NVRAM, and Flash to optimize real-time movement of "hot" data. However, for transfers to SATA-based bulk NFS storage on the back end, customers have the option of performing batch jobs to account for slower disk transfer rates.
"For high-performance to low-performance migrations, Avere uses a batch write mechanism and further optimizes this direction in that only the data that was modified while on the high performance tier needs to be written back to the low performance tier," Tabor said. "Support for solid-state storage is a key element of the Avere architecture, and there is no benefit to batching when migrating between solid-state tiers."
Further discussion and an expert opinion
When presented with competitors' claims, each vendor had counterclaims of their own that are sure to keep this corner of the market interesting. "You have to love the Compellent comparison to Symmetrix V-Max," said Fine, pointing out Symmetrix is a high-end array while Compellent's Storage Center is still focused on the midrange.
Compellent also stripes data across spindles, but "my key message is still the same," Fine said. "By operating inside the volume at the LUN level, we have less data to move, which reduces the need for lots of processors, high-bandwidth internal buses, and large cache."
The EMC spokesperson countered that EMC's midrange array, Clariion, also spreads data over multiple spindles, which means its real-time data migration claim applies to midrange disk array products as well as Symmetrix.
Fine said EMC's claims remain theoretical – while it has shipped LUN-level tiered storage migration with FAST version 1 that was released in December, block-level FAST that integrates with thin provisioning and other advanced features won't be available until later this year.
"EMC doesn't like to compare FAST with Data Progression at all," said Wikibon analyst David Vellante. "But the reality is Compellent has had an automated tiering capability for years."
However, Vellante added, with V-Max's horsepower, if version 2 is delivered as promised, the result could be true real-time data migration.
"Practically speaking, I agree with Compellent that with FAST version 1 most moves will end up, or really should end up, as batch jobs," he said. "FAST version 2 may be another story, but with version 1, my advice to clients would be gain experience with FAST and schedule moves as a batch job. There's really no serious penalty of taking that conservative approach. So it takes a bit longer to tier stuff. Big deal."
Meanwhile, Fine said Compellent is considering adding real-time data progression options. "We get questions about real-time data movement from customers," he said. "We've looked at real-time data movement scenarios and will continue to evaluate implementations for potential adoption."
Dig Deeper on Flash storage for applications