Submitted by Sun CEO Jonathan Schwartz has an interesting blog entry about how Sun Microsystems will start introducing Nand Flash with ZFS as an enterprise storage solution by the end of this year. With the price of Flash memory already plummeting this could be an economical alternative to the expensive NAS solutions.
He’s right in saying storage is going to change a lot in the next few years.
Just a shame his blurb is full of inaccuracies and misinformation.
I’ve always had a dislike for people saying things like “what he wrote was full of inaccuracies” and then not actually telling me what was inaccurate. -1 for you…
Well, I found these sentences rather odd:
“Flash is very fast at reading and writing data, like DRAM […]But unlike either alternative, Flash requires no power to remember data.”
Last time I checked RAM was about 100x faster than Flash and HDDs still had higher burst speeds and didn’t require power to keep data – only for reads and writes.
I guess he could have said “For the typical use case (lots of small, random writes and reads) Flash is faster and more efficient than HDDs”.
Of course, that doesn’t sound nearly as snappy
Doesn’t harddrives require power in order to retain their data over longer periods of time? With flash, you can take a backup and store it for many years and your data is still there, unaffected.
By contrast, any magnetic storage will require you to re-magnetize them now and then, otherwise you’ll lose the ability to distinguish between the bits.
Look at backup tapes for instance, iirc it’s recommended to make use of the tapes atleast once a year, if you don’t, you may lose bits of data over time.
We did a restore to disk of old tapes that hadn’t been used for some 5-6 years and we lost some 20-25% of all the tapes. Even though we used the very same tape unit, the tapes had become unreadable over the years, probably due to not “excercising” the tapes and thus refreshing the magnetized areas.
I still think he shouldn’t have mentioned HDDs and RAM in the same breath. It’s more than just a little misleading since the amount of energy needed to keep the information differs by many orders of magnitude.
Actually, I think he was alluding to the fact that you have to spin up a HDD in order to read data from it and that requires energy. I don’t work in a datacenter but I’d imagine they keep the platters spinning 24/7 to guarantee low latency and avoid wearing out the drives too fast. IIRC, a spin up/spin down is far more harmful to a HDD than running continuously.
Storage is one area of technology infrastructure that has been seriously proprietary. ZFS has in it’s short life already spread out across the BSD’s Mac and possibly even Linux platforms. Nexenta and Pogo Linux have released storage appliances based on the technology. Exciting stuff when compared to the IBM, EMC, NetApp proprietary technologies we have seen to date.
There are two main technologies for flash in ZFS: the ZIL to improve write performance, and the L2ARC to improve read performance (although with current flash-memory based SSDs, we’ll be using the L2ARC to improve random-read performance – and leave sequential read for the disks).
I’ve been busy developing the L2ARC, and recently posted some screenshots of it in action here:
http://blogs.sun.com/brendan/entry/test
These results aren’t the best we can do – this was just to provide an example of it doing something.
A point about power consumption worth mentioning (and that Adam Leventhal has been making), is that the power savings aren’t all about comparing flash-memory based SSDs to disks or DRAM; it is also about allowing slower spinning disks to be used for primary storage, while maintaining high performance through these flash technologies (if you want – the option is there for better price/performance). Check out his ACM article for power consumption numbers:
http://blogs.sun.com/ahl/entry/hybrid_storage_pools_in_cacm