Capacity shortage hits AWS UK micro instances

One of them is at Telecity, Acton... Allegedly.

story from the early days

My previous manager started working with amazon shit back in maybe 2009 I think it was for a company he hired me at a year later. The company in question had very high level connections with amazon to try to resolve the many problems(which didn't help much in the end). Anyway the funny story is on several occasions even in us-east amazon would run out of capacity for large VMs (and maybe others I don't know) at the time and on at least one occasion the customer co-ordinated with amazon on site techs, literally they called and told my manager to log in to the portal with his finger on the button, the amazon people powered up the new hardware and told him when to click the button to allocate resources from the new hardware. In the two and a half years or so I was forced to use amazon cloud I don't recall that situation happening to me(running out of capacity). The customer in question at least on paper at the time peaked spending at about $500,000/month on cloud services(ROI for in house was about 7 months). Because of the executive level connections I don't know how much they actually paid, that was just the size of the bill. This from a tiny shit box startup. Company imploded several years ago.

"Amazon was, at the time of writing, unable to say why its normally elastic compute cloud was unable to keep the newly launched UK micro instances from running dry."

Elasticity isn't magic. If they haven't brought and new hardware quick enough (or there was an abnormally large surge) then there is no hardware and they would have run dry.

Just like real elastic, stretch it too thin and it will snap.

The Cloud...

Other peoples computers they are not building enough of.

What if...

OK, so this is presumably what happens when there's not enough capacity for cheapskates to make use of generous "off peak" pricing for capacity which would otherwise sit idle and earning no revenue, right? In much the same way as a timesharing bureau used to work, right, except the cloud is new an shiny an betterer an HUUUUUUUGE?

What happens when there's a bigger, more serious, capacity shortage for some (global?) reason, and the cheapest cheapskates have already been booted off (not just prevented from starting new stuff)?

That is the way "the cloud" is priced when you look at the small print, right? Everything's fine (==="cheap") till something unusual happens, and then life gets interesting.

Is there a separate (published or unpublished) price for truly "firm" cloud capacity, just as there is for many other things in life?

Same business model as the pallet rental, banking, mobile phone coverage business. Promise lots and lots of people 1% of what you've got; knowing full well they won't all want it at the same time.. (As doumened by Mel Brooks in 'The Producers' ) Ooops

The electricty supply industry solves this problem with mamximum demand meters and variable tarriffs depending on time of day and demand, maybe that's the future.

Not elastic enough?

Is there a reason Amazon can't just automatically subdivide some of their bigger instances into several smaller ones?

TBH, I'd always assumed Amazon worked that way, and would fill their machines with whatever instance sizes people wanted to run. Is there a reason for not doing it that way?

The instances do seem designed to be subdivided - a t2.nano has exactly half the resources of a t2.micro, which has exactly half the resources of a t2.small, which has exactly half the resources of a t2.medium.

Not for prod

Never ever run t2 instances on production environment.