On 11/13/14 15:50, J. Roeleveld wrote:
James,
Backblaze is not selling harddrives. They sell storage on their servers.
The data they collect is based on different drives from different manufacturers.
They are quite open on what they use, check out their website and blogposts.
You should
find the answers to your questions easily there.
Joost
Wow, an honorable vendor, that uses "scientific" standards so other can
readily collect similar data and verify their results. They even put
the scientific method ahead of quarterly profit statements. Impressive.
It's refreshing to know we still have vendors like this. One in a
million. I never suggested they did anything specific like sell HD.
There are a multitude or reasons to publish data that vendor A is better
than vendor B. Furthermore, if an organization at that
size/capitalization is not fully redundant, HA and cost effective, (1)
they wont be in business very long and (2) it really does not matter
because that sort of data only guides your next (corporate) bulk
purchase of hardware, because *price* of component redundant hardware
*always* trumps mtbf specs, imho.
The important thing is for others to be able to independently verify
their results. So the methods, tools and datasets should be readily
available as a first step. Then when IBM, Google and dozens of others,
including but not limited to universities, government labs (Los Alamos)
and private users all collect similar data and discern similar
conclusions, then you know the data is valid and the conclusions are
just. But by then each of the vendors will have new product offerings;
so once again, price drives the market and perceptions adjust the prices
moderately to none at all.
There are a myriad of factors that can affect these results. Independent
verification is the gold standard for data analysis, imho. Not just
"re-crunching numbers" from their datasets, but collecting your own
independent datasets for analysis. I must have missed those links on the
vendor's website? Enhancing the common, open source Network Management
Systems such as nagios or jffnms, where others can readily collect data
on drives themselves, and independently verify their conclusions is
paramount for their published conclusions to hold a "scientific" basis
of validity, imho. Otherwise, it's rather akin to "benchmarks", imho.
Useful ? :: yes ; interesting ? :: most definitely ;
conclusive ? :: not even close!
peace,
James
