Blogger social buttons

Monday, 24 June 2013

How do cloud price drops affect the bottom line? TripAdvisor and Pinterest

Effects of cloud price drops on TripAdvisor and Pinterest

Over the last year and a half, we counted 29 price reductions in cloud services provided by AWS, Google Compute Engine, Windows Azure, and Rackspace Cloud. Price reductions have a direct effect on cloud users, but given the usual tiny reductions, how significant is that effect on the bottom line?

Last year I wrote about cloud cost forecasts for TripAdvisor and Pinterest. TripAdvisor was experimenting with AWS and attempted to process 700K HTTP requests per minute on a replica of its live site, and Pinterest was growing massively on AWS. In the wake of the cloud providers’ price reductions, I revisited TripAdvisor and Pinterest to compare the old and new cloud cost forecasts.

In my first blog post I asked:
  1. How much would it cost to deploy System X on Cloud Y? For example, how much would it cost to host TripAdvisor on the AWS US-East cloud with On-Demand Instances? What if it used one-year Reserved Instances?
  2. What happens to costs when the system grows? For example, Pinterest has around 410TB of data on S3 — what if that keeps growing at a rate of 25 percent every month, as it has done in the last 10 months?

PlanForCloud, a free cloud cost forecasting engine, can help you concentrate on the cost components of your cloud strategy. I created a couple of deployments in PlanForCloud to explore these questions. The results show how important it is update your cost forecasts after price reductions.

Trip Advisor

1. TripAdvisor's "700K requests/minute" deployment on the AWS US-East cloud - how much would it cost?

I first looked at TripAdvisor to try to determine how much its 700K requests/minute on the AWS US-East cloud would cost. I took the following deployment specs from TripAdvisor's blog post detailing its experimental system:

- 270 x Hi-Memory XLarge (m2.xlarge) running 24 hours/day as front-end servers
- 70 x Hi-Memory XLarge (m2.xlarge) running 24 hours/day as back-end servers
- 32 x Hi-Memory Xlarge (m2.xlarge) running 24 hours/day as memcache servers

- 5  x Cluster Compute 8XLarge (cc2.8xlarge) running 24 hours/day as database servers, each having a 1TB EBS volume with 100 IOPS attached

- 5TB of EBS with snapshots to S3 every month for backups of all databases

Data Transfer:
- 145TB of data going out to users every month (around 200GB/hour)

TripAdvisor's deployment on AWS US-East modelled in PlanForCloud

In October 2012, if TripAdvisor used On-Demand instances for the above deployment, they would have expected to pay around $1.7M per year. Re-running the simulation again today, this has been reduced to $1.6M per year due to the price reductions.

If TripAdvisor used one-year Heavy Utilized Reserved Instances, in October 2012 they would have expected to pay around $0.9M per year (including upfront reservation fees). This figure currently stands at $0.7M per year due to the price reductions. The following table shows the percentage reductions for the two scenarios.

One year cost forecast of TripAdvisor

It's important to note that any AWS On-Demand price reductions are passed directly to users, while the upfront cost of Reserved Instances is due at the time of purchase, and it's rare for AWS to issue a partial refund when it reduces this upfront price if it’s past 90 days.


When I wrote my original blog, Pinterest had around 410TB of user data on AWS S3, and historic trends showed that this was growing by around 25 percent every month. In October 2012, Pinterest would have expected to pay around $470K per month after 12 months of growth if they used AWS S3 Standard Storage. Today this figure stands at $362K, thanks to the price reductions. The same amount of user data would have cost around $319K per month if Pinterest used S3 Reduced Redundancy Storage; this figure currently stands at $284K. The following table shows the percentage price reductions. It's interesting to see that S3 Standard Storage has been reduced more than Reduced Redundancy Storage.
Monthly Cost Forecast

Lessons Learned

From these two case studies, it's clear that the significance of cloud price reductions on a business’s bottom line depends on specific scenario and usage levels. PlanForCloud includes more than 12,000 prices across six cloud providers, and these prices change regularly. When you sign-up for PlanForCloud, we'll notify you automatically each time a price reduction affects your cost forecasts.

-- Ali
Technical Lead, PlanForCloud

No comments:

Post a Comment