Reserved Instance Price Reduction for EC2 | Hacker News Reader

Reserved Instance Price Reduction for EC2

aws.typepad.com

33 pointsjeffbarr14 comments

The complexity of trying to figure out how much servers will cost, or should cost, or could cost if I really played with spreadsheets for a few hours-- is what drove me to Rackspace a few years ago. Occasionally someone will tell me that I could save money by going to EC2 and doing something like bidding at a price that's below this or that and something or other... but the whole thing is just annoying to me. Server prices have never been a significant part of my costs compared to other things, so all the game playing is simply tiresome to me. It's simply not practical to project your costs of deploying an application if you use Amazon's hardware and a couple of their services. Am I the only one that feels this way?

I don't know if anyone else feels that way, but when I looked at pricing amazon was cheaper period. Not cheaper if you do X and Y and Z, just cheaper always no matter what. Rackspace was insanely expensive in comparison.

I just started a stopwatch, and decided to try to figure out what you're talking about. After working on an online price calculator with more fields on it than a 1040EZ tax form, and then getting 2/3 of the way through a whitepaper PDF on how pricing works, and trying to figure out what kind of instances I'd need, I stopped at the 10 minute mark. I'm going to go resume work on my software now, but I expect I'm not alone in this.

Maybe I'm just doing it wrong, and there's an easier way to do this. And I'm certain that if I spent $10k a year on servers, I'd take this more seriously.

That just seems like a poor attempt at trolling. "I have to figure out what server I need" is a problem with amazon? I just double checked, and the process is literally identical between amazon and rackspace:

1. pick an instance 2. see the hourly price 3. multiply it by 24 x 30

Which part of this is a problem, and why is it only a problem for you with amazon?

The lesson I've learned from the various reserved instance reductions is to never buy 3-year terms.

I disagree. A 1-year heavy reservation for a Medium High-CPU instance is $765.36. A 3-year for the same is $1515.68, with a per-year amortized cost of $505.22 (the first year will be more expensive than the 2nd and 3rd due to the up-front cost, but we're attempting to minimize the total cost, so I think averaging is fine here). If you got three 1-year reservations, the price reductions required to get the same average cost would have to be in excess of 33%/year. I don't know what the actual price drops have been, but I'd hazard a guess that they haven't been that significant. Is this math off in some way?

I think you're generally right. But there are at least three more factors to consider:

1) the 3-year reservations have a lower per-hour pricing than 1-year

2) the 1-year instance you would purchase a year later will have a lower per-hour price than the 1-year today.

3) you can resell the 3-year instance on amazon's marketplace after you're done with it, so it has residual value.

So it all depends on how much prices drop, what resale values on the "old instances" are, etc.

All in all, the differences probably aren't huge unless you're at massive scale (eg Netflix, Heroku, etc). But at that point you probably negotiate with Amazon directly anyway instead of working off a pricing table.

#1 is the reason why the math works out in favor of 3 year reservations. I took this into account already:

  $price = $reservationCost+$hourlyRate*24*365*$numberOfYears

Where reservation cost, hourly rate, and number of years are pulled off the chart for 1 or 3 year reservations for whatever instance size you're interested in.

#2 is sort of what we're solving for... how much of a yearly price reduction would be necessary to make 3 consecutive 1-year reservations cheaper than a single 3-year reservation (or how much of a 3-year reservation you could effectively abandon and still come out ahead). By my math above, I came up with the magic number being about 40% discount per year making 3x 1-year reservations cheaper than a single 3-year reservation.

  $765 + $765*x + $765*x^2 = $1515
  x = 0.60923

#3 I've yet to use the marketplace, but yes even after discounts happen down the road, your leftover reservation time is certainly worth something, pushing the balance even further in support of 3 year reservations.

The main driver of my comment is that I purchased 3-year terms before Amazon introduced the various categories of utilization. My instances have to run 24/7, so I need "heavy" utilization instances with concommitant higher hourly savings.

By purchasing 3-year RI's prior to the utilization categories, I paid more (350 vs 300) for the reservation than even the costliest new reservation type (heavy) but my hourly rate is higher ($.06 vs $.059) than even the costliest new hourly type (light).

At this point I can't run the numbers, though, b/c I'm not sure how much a 1-year term would have cost me back then. I only know how much I did pay for my 3 year terms.

Maybe my gripe is simply that I feel my pre-category RIs should have been converted to "high" RIs with better hourly rates =)

3-year terms for anything in this space is an eternity. Even if the math checks out, the inflexibility of that kind of lock-in should be a consideration.

Of course, YMMV.

If your needs change you can always list the EC2 Reserved Instance in the Reserved Instance Marketplace.

How liquid is the RI marketplace?

Dunno about other regions, but when I want shopping for instances in EU/Ireland earlier this year I found fuck all.

In a lot of cases you still come out net positive even if you trash the particular service during the latter half of 3 year term.

The lesson is more that you should pay attention to when the last reduction occurred for a given service, and the actual discount you're receiving for reservation. Another good consideration is how much of your time it takes to determine cost analysis against the fact that you're probably investigating cost savings in the tune of +/- 5-15% (EDIT: leaving this here, but this number doesn't make a lot of sense) of your total aws charges anyway.

Reply on news.ycombinator.com