Everywhere I have been recently, the FlashSSD pitchmen have been noisier and noisier. Rob Peglar gave a great cautionary presentation on the subject at the University of Minnesota in Minneapolis a couple of weeks ago, where I also spoke on the agenda of Todd Vojta’s great Paragon Technology Summit.
Peglar noted several things that you should Google before you buy anything Flash.
- “Read Disturb” — the act of reading cells of FlashRAM disturb adjacent cells.
- “Memory wear” — reliability has yet to be determined.
- Non-linear performance — over time or workload
- App blindness — apps and OS don’t use FlashSSD efficiently
- Erase and wait — Cells must be erased before they can be written or rewritten. Erasing takes 100x longer than writing — erase an entire block to write one cell (takes milliseconds, an eon in storage)
This hasn’t stopped everyone and his kid sister from adding FlashSSD as Tier 0 to their arrays (except Peglar’s Xiotech for one). Nimbus Data gave me their pitch the other day and I told them I would write nothing flattering until I had one of their boxes to beat to death. It should be here in July. I will report what I discover.
Meanwhile, here is a bit of historical revisionism that the Flash guys seem to be pushing:
In the beginning, there was the Word.
Eight bits of data written to Flash RAM.
{ 2 comments… read them below or add one }
Jon,
Thanks for bringing to the forefront one of Xiotech’s architectural limitations…which is having a single drive vendor to source from…Seagate.
We’re all certain, when Seagate introduces Flash Drives (they are working on it), Xiotech will jump on the Flash bandwagon and claim that all of the above have been addressed or greatly minimized…
Thanks for this insight Pq65. I am promoting it to a topic so we can all have at it.
You must log in to post a comment.