[Pc_Support] Enthusiasts strike again (oh boy): Samsung MCAQE32G5MPP-0XA (32GB NAND drive)

[Bryan J. Smith]

Here we go again, it seems the enthusiasts are now proliferating the
marketing.  The only negative they see is the price tag (almost $1,000
for these suckers), and _absolutely_ignore_ the limited re-writes that
NAND designs can take.

The "preview," 'Conventional Hard Drive Obsoletism?  Samsung's 32 GB
Flash Drive,' is here:
  http://www.tomshardware.com/2006/09/20/conventional_hard_drive_obsoletism/  

In typical enthusiast fashion, they make several statements that are
just dead _false_.

1. 'Flash hardware has intrinsic benefits, as it benefits from
    extremely short access times, but it also has specific
    advantages for the upcoming Windows Vista operating system.
    Flash memory as an optional cache allows the user to take
    advantage of Vista's "ReadyBoost" feature, allowing for a
    peppier PC.'

No, the "ReadyBoost" feature is one way Vista can use NAND EEPROM
with_out_ killing the life of the unit as a standard hard drive.  It
relocates _read-only_ data -- such as boot-time and application binaries
-- things that don't change.  It _prolongs_ the device's lifetime.

2. 'In terms of our test for drive interface bandwidth the limiting
    factor is the UltraATA/66 interface effectively handicapping
    scores, especially in the data read category.'

No, wrong again.  If you read the engineering specification sheet for
the K9WAG08U1M-PCB0 (2Gx8b), the _read_ DTR is 2KiB every 20us
(microseconds) from the serial NAND gate.  That's basically a limit of
100MBps "ideal."  So the reason for an UltraDMA mode 4 (66MBps)
interface is because it can't break 50MBps. 

http://www.samsung.com/Products/Semiconductor/NANDFlash/SLC_LargeBlock/16Gbit/K9WAG08U1M/ds_k9xxg08uxm_rev10.pdf

Now NAND EEPROM _does_ give _consistent_ "burst" DTRs.  In the charts,
you can see it gives _solid_ and _consistent_ performance throughout
many operations -- as long as they are heavy _reads_.  But once you dip
into lots of random modifications -- like databases -- well, it's all
over.  ;->

Now one might want to point out the "lifetime" in the specs as 100,000
writes.  That is for the _entire_ 2Gx8b device.  It has spare cells that
are reallocated as they fail, and it assumes a good amount of
randomization -- some of which can be done by the device, although the
OS needs to help.  The NAND CMOS cells themselves fail at sub-1,000
writes each.

It's great to see the technology coming.  But it won't be long into
their commoditization that people will stand back and go, "why do these
things fail so often?"  Hopefully many of you will remember my posts
when you start to hear that.

Or maybe by that time, most of the OSes will be true read-only boots and
not assume they can write anytime, anywhere.  Or at least there will be
hybrid NAND EEPROM and disc so NAND EEPROM is used effectively as it is
intended -- write a few times, read many.  That's where both the
performance and lifetime excel.

But NAND EEPROM is likely to _never_ replace generic disc as the
secondary storage of choice.  3D crystal-laser is far more likely to do
that in the next decade, maybe sooner.


-- 
Bryan J. Smith          Professional, technical annoyance
mailto:b.j.smith at ieee.org    http://thebs413.blogspot.com
---------------------------------------------------------
The world is in need of solutions.  Unfortunately, people
seem to be more interested in blindly aligning themselves
with one of only two viewponts -- an "us v. them" debate
that has nothing to do with finding an actual solution.