Q&A: The benefits of the spot instances market

For companies looking to save money on cloud computing costs, Amazon’s spot instances market is both a blessing…and a curse. While its “spot” market can deliver savings that are 90 per cent lower than on-demand cloud computing costs, when spot availability is full, not only can pricing fluctuate wildly, capacity is not always guaranteed. This can leave cloud customers with interruptions in services as they switch to another server.

Spotinst, an Israel-based startup, is one of several companies attempting to overcome this challenge. We spoke with co-founder and CEO Amiram Shachar about the current state of the spot instances market.

1. What is the spot instances market and why is it important?

Amazon’s EC2 (Elastic Compute Cloud) “Spot Market” allows Amazon Web Services customers to name their price for compute capacity, which can be as much as 90 per cent cheaper than the same infrastructure offered through the On-Demand business model.
The AWS EC2 Spot Market is made up of hundreds of smaller compute markets with real-time supply and demand-based pricing. These markets are made up of unused AWS capacity, where the price - determined by market forces - is less than 1 cent per core hour using the latest generation Intel chips. At this price point we fundamentally believe the total cost of ownership for cloud compute is materially cheaper than even the most efficient enterprise data centres.

2. What kind of companies should be leveraging the benefits of the spot instances market?

Every company who uses AWS can and should leverage the spot market as it helps them achieve more efficiency as well as reduce EC2 costs. The spot market offers the opportunity to significantly lower the cost of compute, accelerate parallel processing and fundamentally shift the economics of high throughput processing.

3. What are the limitations of the spot instances market?

There are 3 main limitations when using the spot instances market:
A) SLA – As opposed to standard contracts, the service level agreement in the spot instances market does not guarantee that capacity will be available. In fact, instances might be terminated from time to time without advanced notice, forcing customers to switch servers in the middle of a project.

B) Pricing – The spot instances market does not afford users a predictable or fixed pricing structure and changes based on supply and demand. Prices can be anywhere from 90 per cent cheaper to up to 10x higher than the regular server price. Additionally, as the demand for spot instances has grown, it has become increasingly difficult for customers to secure the capacity they need at affordable prices.

C) Safe application draining – Prior to spot termination, the customer must be sure to safely shutdown all applications that they are running. Additionally, they must save all logs and close active connections, etc...)

4. Who are the main players in the spot instances market?

There are several big spot instances consumers, including numerous universities and research labs. The most well-known spot instances consumer, however, is most probably Netflix, which is such a large spot instances consumer that it has created its own internal spot market and now coaches other companies on how they can do the same. In a blog post detailing the creation of its internal spot market, Neflix wrote that “in the first month of deployment, we observed significant benefits in terms of performance and throughput. We were successful in making use of Netflix idle capacity for production, research, and QA. Our encoding velocity increased dramatically. Experimental research turn-around time was drastically reduced. A comprehensive full regression test finishes in half the time it used to take. With a cost efficiency of 92 per cent, the spot market is not completely free but it is worth the cost.”

5. How do companies bid on spot instances?

The spot instances bid process begins when users create a spot instance request or a spot fleet request. According to AWS, the bid price is defined by the maximum price that users are willing to pay per hour per instance. It also includes other constraints such as the instance type and Availability Zone. As outlined on the AWS site, if the bid price is greater than the current spot price for the specified instance, and the specified instance is available, the request is fulfilled immediately. Otherwise, the request is fulfilled whenever the spot price falls below the bid price or the specified instance becomes available. Users have different strategies when it comes to bidding on spot instances. Some bid a price lower than the on-demand price, while others might bid higher (to gain more availability).

6. Is there a typical spend on spot instances?

While the typical savings are 85 per cent off the regular price, there are no such things as typical spends as each case depends on the company and the deployment scenario.

7. Are spot instances right for every company? If not, why not?

Yes, any company which regularly uses cloud computing can stand to significantly benefit from using the spot instances market. Among our users, we have customers from a range of industries, including mobile forensics, app monetisation, gaming and retail companies.

8. What do you think the future of the spot instances industry will look like? Will we continue to see consolidation along the lines of the ClusterK acquisition?

The compute market worldwide is heading more and more towards 'market driven' pricing, with cloud infrastructure services companies (e.g. Amazon, Microsoft, Google, IBM) continuing to add more capacity (approximately 50 per cent year over year). While the demand for compute has grown, there is still more supply than demand. This has forced cloud providers to incentivise their customers to purchase unused capacity, or suffer losses. On the other hand, enterprises and startups will keep looking for cost-optimised solutions they can use to grow and scale their cloud computing deployments.

We believe that Infrastructure as a Service (IaaS) offers them the best set of tools to help them efficiently scale their cloud deployments. In a sentence, IaaS is a form of cloud computing that provides virtualised computing resources over the Internet.