GCN LAB IMPRESSIONS

Flash memory in government data centers? The pros and cons.

It seems we’ve been talking about using flash memory and solid state drives instead of traditional spinning storage media for many years. But the announcement that IBM has purchased Texas Memory Systems, primarily to get access to its flash technology, raises the possibility of having flash technology move into federal data centers in the near future.

But is this something that feds should want?

As of late, IBM has been experimenting with memory and storage, including some really futuristic stuff like the liquefied memory Greg Crowe reported on last year. Given that the two primary concerns inside data centers are storage and memory, it makes sense.

There are a ton of advantages to flash memory, plus a few pitfalls. Let’s take a look at the pros and cons.

On the plus side, there are no moving parts, so it’s almost never subject to any type of breakdown, certainly not of the mechanical sort. And flash is rugged, so it can survive unusual events like earthquakes or some new techie bumping into the RAID storage cabinet.

Disk access is also fast. Without any arms or spinning parts, flash storage is almost like RAM in that everything is handled electronically. In fact, the way to build a supercomputer these days is to use flash drives for the extra zip they provide, as Lawrence Livermore labs proved a few years ago.

Another advantage that is often not considered is that a flash drive generates very little heat compared to its mechanical counterparts. Multiply that by thousands of drives in a data center and you are talking about a huge energy savings just trying to keep the place cool.

But there are disadvantages, mainly two of them.

First is price. Although flash drives are coming down in price at some levels —companies are just giving away those little promotional flash key drives — they remain expensive for large-capacity drives of the type you would find in a data center. Or at least they are more expensive than non-solid state drives. But that’s not to say that can’t change. Remember when tape backups were the top of the line before mechanical drives got their act together? IBM’s move to get hold of Texas Memory Systems might help bring large-capacity flash prices down a bit.

The other negative is that flash memory can’t be written to forever. There is a physical limit to the number of times that data can be written to or pulled from it. However, in recent years this limit has been raised to millions of writes. So although a write-to limit still technically exists for flash drives, in terms of the real world, a mechanical drive is also likely to fail sometime during a million-write lifetime. Even though it may not have an official expiration date, its mechanical parts will wear out. We just don’t know exactly when.

I’ve seen data center solutions where frequently accessed data is put onto flash drives while long-term storage is handled by mechanical drives. This speeds up common access tasks, yet keeps the less-used data stored on cheaper drives. That is a great interim way to build a data center.

But now could be the time to move to all flash, especially if IBM can lower the prices on the biggest drives to the point where it becomes a wash when you factor in lower environmental costs, quicker access to data and better overall reliability and uptime.

So yes, feds should probably start thinking about changing their data centers over to all flash, or at least considering it should they happen to be close to upgrading, or adding new data centers to their agencies.

About the Author

John Breeden II is a freelance technology writer for GCN.

Reader Comments

Wed, Aug 22, 2012

The key to this, as in all backup systems, is redundancy. The technical variants that Ric Halsaver mentions in his comment are well received, but to reduce the probability of data loss you simply have redundancy built in, preferably at different locations.

Wed, Aug 22, 2012 Ric Halsaver Wash DC

To your comments of the type of flash chip, it makes sense to put eMLC as the low-end flash, as it permits 30,000 writes per bloxk as opposed to the 3-5,000 that MLC provides. This will allow the datacenter to overcome chip write exhaustion. For heavy writes, they sould use SLC and auto tiering. Ric Halsaver, VP Federal Systems, Texas Memory Systems

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above