Can You Scale a Cockroach?
by David Merrill on March 11, 2010
I am working in India this week, and in meetings with a new colleague from Infosys we got on the topic of compute scaling. I was explaining the situations that we see emerging with JBOD storage for some new cloud and email architectures, and the principle of “what works small may not work large.”
He then told me of this website and how the concept of scaling an insect (like a cockroach) does not work in SciFi movies. The cockroach is perfectly engineered for its size and function, but if you were to enlarge the cockroach by 50 or 100 times, it would be crushed by its own weight.
After the meeting, I could not wait to get this website, and the discussion on the square cube law.
Thinking about this more on the flight from Mumbai to Tokyo, these principles do apply to storage architectures. I am working on some econ models of JBOD for exchange 2010, and to demonstrate the inflection point (economically) of when these JBOD designs fall apart at scale. I may have referenced this wikibon article previously on SAN vs. DAS for exchange, and others seem to agree that some storage designs work well small, but should not be expected scale.
Lets keep those cockroaches to their normal size now…
Comments (2 )
John Winney on 15 Mar 2010 at 8:37 am
Fantastic. At last the square cube law let’s us put giant ants in our storage presentations!
Joking aside, this is a very nice analogy. Thanks Dave, its gems like this I like.
Martin Scholl on 31 Mar 2010 at 9:59 am
Hello David,
obviously you are right in your cockroach-analogy. But nature also has a good answer to it: if a large-scale ant can’t handle it, many small ants will. In other words: take the most efficient tool you can find and _combinate_ it to do the job.
Actually this is what “those cloud/JBOD guys” do — and it just works.
Martin
Disclosure: I work for one of “those cloud/JBOD guys”.
