|
|
Calling for an End to Unrealistic SSD vs HDD IOPS
Comparisons |
Editor:- I've been publishing
articles about the
SSD market for over a
decade -
so maybe that's why something in my brain snapped yesterday - when I was reading
yet
another
new article from an SSD vendor about the server acceleration which you could
get from their product.
Because the only competition it mentioned was
hard drives.
Enough
is enough! - I said.
I can't remember whether the article claimed
their product was 300x or 500x faster in random IOPS than a 15K
SAS hard drive -
but as the IOPS in a hard drive haven't changed much since the appearance of the
first 15K RPM products in
EMC's CLARiiON
systems in 2002, and as there may
never
be a 20K RPM drive... The long and the short of it is that every month the
fastest SSDs get
better - while the fastest hard drives remain exactly as fast as they were. So
the HDD versus SSD random IOPS gap gets wider. We haven't learned anything
new!
I'm not denying that these comparisons have been useful in the
past. And they are still useful when a single
2.5" SSD for
example is being compared with a single 2.5" hard drive. But as I wrote in
an article in
2003 when you're looking at what a fast
rackmount SSD can
do for you - you should be calculating how many servers you can save /
or would need to add to get to the same 3x application speedup - not how
many disks.
It's understandable that people refer new products to an
assumed common frame of reference.
I'm sure that a century ago when
the first car owners were talking to their neighbors - they must have made
comparisons like - how much faster they were than a horse.
"It's
amazing. Unlike my old horse my new car doesn't get tired. And I don't need
pasture or hay to make it go - just a few bottles of gas."
And
you didn't get the same stinking mess in the city streets either. (At the time
the Sherlock Holmes stories were being published in the Strand Magazine in 1892
- there was a "green" market carting horse dung out of London to stop
the streets becoming impassable.)
I think I'm safe in assuming that the
last time you bought a car - you didn't compare its speed or fuel consumption
to a horse. Am I right?
No - of course not. You compared it to other
cars.
It's a sign of a maturing market when there are enough products
around to compare them to each other, instead of comparing them to what came
before. And we have reached that time with SSDs. It's taken
30 years
- but that's another story.
This new wave of comparing SSDs to each
other kicked off with the article
RAM SSDs versus
Flash SSDs - which is Best? - which included views from leading SSD
oems on both sides of the fence. And other articles have demolished the
myths about SLC
flash SSDs while highlighting the
risks of using
MLC flash in inappropriate server applications.
In my view
comparing SSDs to HDDs does not give you a useful picture when you're looking
at options in the server acceleration market. |
 |
Although I expect that such
weak comparisons will be sprinkled in lite weight SSD articles for the next
few years - it's time for some stronger seasoning. It's time for more articles
by SSD vendors to say how their products and technologies compare with other
SSDs. | |
|
| Footnote:- here are
more details about the specific vendor article which triggered my pleading
above.
| |
|
|
| Here are some popular SSD
articles / features.
|
| . |
| Z's Laws - Predicting
Future Flash SSD Performance |
A reader asked me a
very good question.
"Is there an industry roadmap for future
flash SSD
performance?"
That prompted other questions like...
- How fast are flash SSDs going to be in 2009?, 2010? or 2012?
- What are the technology factors which relate to flash SSD throughput and
IOPS?
- How close will flash SSDs get to
RAM SSD performance?
There wasn't a simple answer I could give at the time. Clues lay
scattered all across this web site
and in my many one on one discussions with readers about the market... |
 |
But I agreed there should be
a single place on the web where these answers could be found.
Forget
Moore's
Law. That gives you the wrong answer, and this article explains why. ...read the article | | | |
