2 Comments

Summary:

Publishing analytics startup Parse.ly just switched its entire IT footprint to Amazon Web Services, saving money and improving performance in the process. As clouds keep getting cheaper and better, resistance to them is becoming futile.

Inside a Google data center. Image courtesy of Google
photo: Google

Resistance to cloud computing might not be futile, but it’s at least beginning to look foolish — especially as services from the top providers such as Amazon Web Services keep getting cheaper while their performance gets better. It’s also looking like smaller-scale or “enterprise” cloud platforms will have to promise some serious differentiation in order to justify their higher costs.

To highlight this trend, here’s a chart from publishing analytics startup Parse.ly graphing its IT spending from inception until early 2014.

server_costs

 

The long story made short — but you can read the whole thing up through September 2013 here — is that Parse.ly started off using Rackspace primarily and AWS for backup and a variety ad hoc workloads (e.g., Hadoop jobs). In 2012, it opted to cut costs by switching its primary analytic database to physical servers in a co-location center while continuing to run its cloud workloads primarily in Rackspace. In late 2013, it began transitioning more workloads to AWS and completed an entire transition to AWS in late February 2014.

After paying double (to both Rackspace and AWS) during the transition, Parse.ly is now paying less than monthly than it was before making the move. Its spending patterns might be unique because of the workloads it’s running, but they’re compelling nonetheless.

And, Parse.ly Co-founder and CTO Andrew Montalenti told me, there’s icing on this cake, as well: “What’s crazy is we got a speed-up and saved money.” The company’s primary analytics database is now running significantly faster on AWS SSD-backed instances than it did on bare metal (albeit hard-disk-backed) servers.

If recent claims from Google about adjusting its pricing in accordance with Moore’s Law come true — and if its unique strategy around price reductions on long-running instances catches on — we should be in for continually lower prices on basic cloud computing services. AWS is the cloud king, but Google and Microsoft are positioned as strong contenders, and if low costs are what wins users, they’ll all play along to ensure no one else owns that story. The same goes for improved performance and rapid feature updates.

More and more, it looks like the future of cloud computing will be renting the infrastructure that lets users operate like, well, Amazon, Google and Microsoft but at a fraction of the cost (and scale). We’ll hear a lot more about where the industry is heading at our Structure conference, which takes place June 18 and 19 in San Francisco, and features, among many others, Google’s Urs Hölzle, Amazon’s Werner Vogels and Microsoft’s Scott Guthrie.

  1. Thanks Derrick for an excellent article – I hope this sparks discussion! I would have to contradict with the arguments made – while scale is important, it is far from the only thing enabling lower prices.

    With the information provided, I’ll try and understand the true scale of Parse.ly’s infrastructure. If they run AWS’ on-demand servers (i2.4xlarge -instance with the specified resources priced at $3.751 per hour) a single instance would cost about $2700 a month (30 days/720 hours). While the monthly costs are at around $20 000 for the complete infrastructure, this would give Parse.ly about 7.4 full instance months in size.

    This would roughly translate the size of the infrastructure to about 118 CPUs, 902GB of memory and about 23680 GB of SSD.

    Now, I’m pretty sure Parse.ly runs a wide set of services in a variety of ways, not limited to different investments in AWS’ reserved instances and so forth, and therefore such a simple calculation would not justify the most honest comparison, but for the sake of my argument – let’s run with this.

    For the above mentioned infrastructure Parse.ly pays AWS about $20 000 a month.

    I represent UpCloud, a pure IaaS provider out of Finland and it is exactly in these kind of cases that the likes of us (there are multiple other relatively similar providers out there) are able to beat AWS, Rackspace and others – the big guys which are thought to be offering the best deals to the market.

    On UpCloud – Parse.ly would pay for the similar resources at the following rates (our London data centre, we will have similar pricing in Chicago very soon):

    118 CPUs x 7.20€/CPU/month = 849.60€
    902GB of memory x 3.60€/GB/month = 3247.20€
    23680GB of SSD x 0.20€/GB/month = 4736€

    In total: 8832.80€ a month. With today’s rates Google converts this to about $12 223.

    So out of the box UpCloud would save Parse.ly about $8000 or 40% of their costs – would they only use AWS on-demand pricing. Traffic is not calculated into the equation at all, but since we charge 0.05€ per outbound GB and hence this is also lower than AWS’ pricing – it does not make sense to calculate it in.

    As said, you can optimize costs with AWS’ reserved instances, but with such decisions you’re already committing yourself to fixed terms, which go against the cloud philosophy of being always agile and scalable – hence this simplification for the sake of the argument.

    Long rant, but what I’m trying to articulate here with an extreme simplification – is that while nobody gets fired for buying IBM… I mean AWS, it’s not the best alternative for everyone. Smaller IaaS providers like us, we are extremely passionate about the way they build our technology. With these decisions at the core of how we function – we are able to offer better prices for similar services from AWS.

    Comparing IaaS providers is a tough game, but scale isn’t the only thing that is able to bring down prices. It’s innovation and technological elegance combined with rock solid experience. That capability also lives in companies like ours, hence the very concrete price advantage to larger players like AWS.

    Antti Vilpponen
    UpCloud
    http://upcloud.com

    Reply Share
  2. Respectfully, I stopped reading at “for the sake of my argument”. Your argument is just that – an argument, not reality.

    Reserved Instances is not a type of AWS server. It’s simply a billing construct so there’s zero functional difference between a Reserved Instance of On-Demand instance. You can apply an RI to an On-Demand instance and the discounting begins within the hour to click to purchase.

    Hence, there is zero sense in using AWS as your IaaS and not including RIs as a core component of reducing your bill. Compare apples to apples.

    I agree, scale isn’t the only thing. Amazon also was built on a successful retail, customer obsessed culture and this bleeds into their web services division. Combine that with Amazon’s massive lead-time and massive cloud and this result thins the air so you essentially run out of breath.

    If you don’t have a sizable war chest like MSFT or GOOG, good luck to you. AWS is here for the long haul.

    Reply Share