Calling the Shots with Flash, Disk, and Tape
June 3, 2014 Dan Burger
IBM‘s investment in flash storage is an indication Big Blue believes enterprise-level computing is ready for flash. One of its moves is a partnership with Tributary Systems, a company that produces a high performance virtual tape library product called Storage Director. Tributary has just released a virtual backup application called flash tape, which combines Storage Director and IBM FlashSystem storage arrays.
IBM FlashSystem emphasizes performance and microsecond response times with reductions in transaction times, batch processing times, and energy consumption efficiencies. Flash tape combines speed and utility–like the creation of a 250 mph Ford Econoline van. Tape has reliability (check the facts, not the FUD), capacity, and is a growth market in enterprise computing.
Storage Director can do any-to-any virtual tape backup solution for IBM and non-IBM servers. Its benefit is in simultaneously backing up data at high speed from a variety of heterogeneous servers and then writing the data per pre-set rules to multiple storage technologies, which includes flash.
The purpose of flash tape is to allow companies to access the data they want and put it in the device where it makes the most sense, whether that is for quick access or for archival retrieval, while simultaneously creating a disaster recovery plan.
“Companies can put data on flash as the immediate target when application access to that data needs to be lightning quick,” says Ed Ahl, director of business development at Tributary. “They can hit the flash to pull data in and out of applications. And once it is on flash, they can also make a secondary copy to physical tape for a disaster recover copy, without making a secondary move.”
Using the FlashSystem 820 flash appliance, Ahl says Tributary tested its flash-tape capability running two streams, a total of 8 GB, pushing data at the rate of 2.5 GBs per second.
Storage Director is software that runs on IBM System x servers and looks like a 3584 L32 virtual library to Power Systems boxes running IBM i.
According to Ahl, Storage Director is capable of pushing data to flash and backing it up to tape faster than it could push data to a hard drive system.
“For us, the limitations have always been hard drives,” Ahl says. “On an 8 GB Fibre to a disk system, we can hit 750 MBs per stream. And with flash, there are no latencies. Finding high performance RAID arrays has been a bottleneck in the past. That is not a problem with Flash.”
Tributary is testing de-duplication capabilities so that Storage Director can dedup back-end data to the Flash storage. Ahl believes performance will be noticeable, given that dedup typically robs systems of speed.
“Storage Manager can pull in data faster than any deduplication device,” Ahl claims. “We used to run dedup as a back-end process, but now we will take a flash box and make it competitive with any disk system. We also do LZ compression and encryption on the fly. And we are seeing compression rates between 5x and 15x on IBM i. We take a 20 TB flash unit and make it look like 100 TB.”
Inside Storage Director is a media manager consisting of two databases. One is for incoming data and the other is a “marker” base, which identifies where the data resides on the tape. If data is no longer in flash storage when the application needs to access it, the marker base loads and streams the tape to the location of the data. It then reads it from the tape (not form cache). When running on LTO 6 media, Ahl claims about 350 MB/sec throughput. Not as quick as flash cache, but much faster than what users are expecting.
“Take into account what it costs to run high performance disk arrays, including power consumption,” says Levi Norman, senior marketing manager for IBM FlashSystems. “Compare that with a flash array running a database that is one to two terabytes. The cost comes out in favor of Flash.”
Norman noted a company called Plenty of Fish, an online dating site, which has more than 80 million users and is using all flash storage–no disks. Plenty of Fish claims it would have to double or triple the number of servers it uses if it relied on disk arrays and the licensing costs would be unaffordable.