The Cost of Cloud Computing Is, in Fact, Dropping Dramatically

The Cost of Cloud Computing Is, in Fact, Dropping Dramatically

You might read the headline statement that the cost of cloud computing is dropping and say “Well, duh!”. Or maybe you’re on the other side of the fence. A coworker recently referred me to a very interesting blog on the Kapwing site that states Cloud costs aren’t actually dropping dramatically. The author defines“dramatically” based on the targets set by Moore’s Law or the more recently proposed Bezos’ Law, which states that “a unit of [cloud] computing power price is reduced by 50 percent approximately every three years.” The blog focused on the cost of the Google Cloud Platform (GCP) n1-standard-8 machine type, and illustrated historical data for the Iowa region:

DateN1-standard-8 Cost per Hour
January 2016$0.40
January 2017$0.40
January 2018$0.38

The Kapwing blog also illustrates that the GCP storage and network egress costs have not changed at all in three years. These figures certainly add up to a conclusion that Bezos’ Law is not working…at least not for GCP.

Whose law is it anyway?

If we turn this around and try to apply Bezos’ Law to, well, Bezos’ Cloud we see a somewhat different story.

The approach to measuring AWS pricing changes needs to be a bit more systematic than for GCP, as the AWS instance types have been evolving quite a bit over their history. This evolution is shown by the digit that follows the first character in the instance type, indicating the version or generation number of the given instance type. For example, m1.large vs. m5.large. These are similar virtual machines in terms of specifications, with 2 vCPUs and about 8GB RAM, but the m1.large was released in October 2007, and the m5.large in November 2017. While the “1” in the GCP n1-standard-8 could also be a version number,  it is still the only version I can see back to at least 2013. For AWS, changes in these generation numbers happen more frequently and likely reflect the new generations of underlying hardware on which the instance can be run.

Show me the data!

In any event, when we make use of the Internet Archive to look at  pricing changes of the specific instance type as well as the instance type “family” as it evolves, we see the following (all prices are USD cost per hour for Linux on-demand from the us-east-1 region in the earliest available archived month of data for the quoted year):

m1.largem3.largem4.largem5.largeReduction from previous year/generation3-year reduction
2008$0.40
2009$0.400%
2010$0.34 -18%
2011$0.340%-18%
2012$0.32-6%-25%
2013$0.26-23%-31%
2014$0.24$0.23-13%-46%
2015$0.175$0.14-64%-103%
2016$0.175$0.133$0.120-17%-80%
2017$0.175$0.133$0.108-11%-113%
2018*$0.175$0.133$0.100$0.096-13%-46%

*Latest Internet Archive data from Dec 2017 but confirmed to match current Jan 2018 AWS pricing.

FWIW: The second generation m2.large instance type was skipped, though in October 2012 AWS released the “Second Generation Standard” instances for Extra Large and Double Extra Large – along with about an 18% price reduction for the first generation.

To confirm that we can safely compare these prices, we need to look at how the mX.large family has evolved over the years:

Instance typeSpecifications
m1.large (originally defined as the “Standard Large” type)2vCPU w/ECU of 4, 7.5GB RAM
m3.large2vCPU w/ECU of 6.5, 7.5GB RAM
m4.large2vCPU w/ECU of 6.5, 8GB RAM
m5.large2vCPU w/ECU of 10, 8GB RAM

A couple of notes on this:

  • ECU is “Elastic Compute Unit” –  a standardized measure AWS uses to support comparison between CPUs on different instance types. At one point, 1 ECU was defined as the compute-power of a 1GHz CPU circa 2007.
  • I realize that the AWS mX.large family is not equivalent to the GCP n1-standard-8 machine type mentioned earlier, but I was looking for an AWS machine type family with a long history and fairly consistent configuration(and this is not intended to be a GCP vs AWS cost comparison).

The drop in the cost of cloud computing looks kinda dramatic to me…

The net average of the 3-year reduction figures is -58% per year, so Bezos’ Law is looking pretty good. (And there is probably an interesting grad-student dissertation somewhere about how serverless technologies fit into Bezos’ Law…)  When you factor the m1.large ECU of 4 versus the m5.large ECU of 10 into the picture, more than doubling the net computing power, one could easily argue that Bezos’ Law significantly understates the situation. Overall, there is a trend here of not just a significantly declining prices, but also greatly increased capability (higher ECU and more RAM), and certainly reflecting an increased value to the customer.

So, why has the pricing of the older m1 and m3 generations gone flat but is still so much more expensive? On the one hand, one could imagine that the older generations of underlying hardware consume more rack space and power, and thus cost Amazon more to operate. On the other hand, they have LONG since amortized this hardware cost, so maybe they could drop the prices. The reality is probably somewhere in between, where they are trying to motivate customers to migrate to newer hardware, allowing them to eventually retire the old hardware and reuse the rack space.

Intergenerational Rightsizing

There is definite motivation here to do a lateral inter-generation “rightsizing” move. We most commonly think of rightsizing as moving an over-powered/under-utilized virtual machine from one instance size to another, like m5.large to m5.medium, but intergenerational rightsizing can add up to some serious savings very quickly. For example, an older m3.large instance could be moved to an m5.large instance in about 1 minute or less (I just did it in 55 seconds: Stop instance, Change Instance Type, Start Instance), immediately saving 39%. This can frequently be done without any impact to the underlying OS. I essentially just pulled out my old CPU and RAM chips and dropped in new ones. Note that it is not necessarily this easy for all instance types – some older AMI’s can break the transition to a newer instance type because of network or other drivers, but it is worth a shot, and the AWS Console should let you know if the transition is not supported (of course: as always make a snapshot first!)

Conclusion

For the full view of cloud compute cost trends, we need to look at both the cost of specific instance types, and the continually evolving generations of that instance type. When we do this, we can see that the cost of cloud computing is, in fact, dropping dramatically…at least on AWS.

Cloud Per-Second Billing – How Much Does It Really Save?

Cloud Per-Second Billing – How Much Does It Really Save?

It has been a little over a month since Amazon and Google switched some of their cloud services to per-second billing and so the first invoices with the revised billing are hitting your inboxes right about now. If you are not seeing the cost savings you hoped for, it may be a good time to look again at what services were slated for the pricing change, and how you are using them.

Google Cloud Platform

Starting with the easiest one, Google Cloud Platform (GCP), you may not be seeing a significant change, as most of their services were already billing at the per-minute level, and some were already at the per-second level. The services moved to per-second billing (with a one-minute minimum) included Compute Engine, Container Engine, Cloud Dataproc, and App Engine VMs.  Moving from per-minute billing to per-second billing is not likely to change a GCP service bill by more than a fraction of a percent.

Let’s consider the example of an organization that has ten GCP n1-standard-8 Compute Engine machines in Oregon at a base cost of $0.3800 per hour as of the date of this blog. Under per-minute billing, the worst-case scenario would be to shut a system down one second into the next minute, for a cost difference of about $0.0063. Even if each of the ten systems were assigned to the QA or development organization, and they were shut down at the end of every work day, say 22 days out of the month, your worst-case scenario would be an extra charge of 22 days x 10 systems x $0.0063 = $1.3860. Under per-second billing, the worst case is to shut down at the beginning of a second, with a highest possible cost for these same machines (sparing you the math) being about $0.02. So, the best this example organization can hope to save over a month with these machine with per-second billing is $1.39.

Amazon Web Services

On the Amazon Web Services (AWS) side of the fence, the change is both bigger and smaller.  It is bigger in that they took the leap from per-hour to per-second billing for On-Demand, Reserved, and Spot EC2 instances and provisioned EBS, but smaller in that it is only for Linux-based instances; Windows instances are still at per-hour.

Still, if you are running a lot of Linux instances, this change can be significant enough to notice.  Looking at the same example as before, let’s run the same calculation with the roughly equivalent t2.2xlarge instance type, charged at $0.3712 per hour. Under per-hour billing, the worst-case scenario is to shut a system down even a second into the next higher hour. In this example, the cost would be an extra charge of 22 days x 10 systems x $0.3712 = $81.664. Under per-second billing, the worst case is the same $0.02 as with GCP (with fractions of cents difference lost in the noise). So, under AWS, one can hope to see significantly different numbers in the bill.

The scenario above is equally relevant to other situations where instances get turned on and off on a frequent basis, driving those fractions of an hour or a minute of “lost” time. Another common example would be auto-scaling groups that dynamically resize based on load, and see enough change over time to bring instances in and out of the group. (Auto-scale groups are frequently used as a high-availability mechanism, so their elastic growth capabilities are not always used, and so savings will not always be seen.) Finally, Spot instances are built on the premise of bringing them up and down frequently, and they will also enjoy the shift to per-second billing.

However, as you look at your cloud service bill, do keep in mind some of the nuances that still apply:

  • Windows: GCP applies per-second billing to Windows; AWS is still on one-hour billing for Windows.
  • Marketplace Linux: Some Linux instances in the AWS Marketplace that have a separate hourly charge are also still on hourly billing (perhaps due to contracts or licensing arrangements with the vendors?), so you may want to reconsider which flavor of Linux you want to use.
  • Reserved instances: AWS does strive to “use up” all of the pre-purchased time for reserved instances, spreading it across multiple machines with fractions of usage time, and per-second billing can really stretch the value of these instances.
  • Minimum of one-minute charge: Both GCP and AWS will charge for at least a minute from instance start before per-second billing comes into play.

Overall, per-second billing is a great improvement for consumers of cloud resources…and will probably drive us all more than ever to make each second count.

AWS IAM Roles and Ways to Use them to Improve Security

AWS IAM Roles and Ways to Use them to Improve Security

What are AWS IAM Roles?

Within AWS Identity and Access Management system (IAM) there are a number of different identity mechanisms that can be configured to secure your AWS environment, such as Users, Groups, and AWS IAM Roles. Users are clearly the humans in the picture, and Groups are collections of Users, but Roles can be a bit more obscure. Roles are defined as a set of permissions that grant access to actions and resources in AWS. Unlike Users, which are tied to a specific Identity and a specific AWS account, an IAM Role can be used by or assumed by IAM User accounts or by services within AWS, and can give access to Users from another account altogether.

To better understand Roles, I like the metaphor of a hat.  When we say a Role is assumed by a user – it is like saying someone can assume certain rights or privileges because of what hat they are wearing.  In any company (especially startups), we sometimes say someone “wears a lot of hats” – meaning that person temporarily takes on a number of different Roles, depending on what is needed. Mail delivery person, phone operator, IT support, code developer, appliance repairman…all in the space of a couple hours.

IAM Roles are similar to wearing different hats this in that they temporarily let an IAM User or a service get permissions to do things they would not normally get to do.  These permissions are attached to the Role itself, and are conveyed to anyone or anything that assumes the role.  Like Users, Roles have credentials that can be used to authenticate the Role identity.

Here are a couple ways in which you can use IAM Roles to improve your security:

EC2 Instances

All too often, we see software products that rely on credentials (username/password) for services or accounts that are either hard-coded into an application or written into some file on disk. Frequently the developer had no choice, as the system had to be able to automatically restart and reconnect if the machine rebooted, without anyone to manually type in credentials during the rebootwhen the system rebooted. If the code is examined, or file system is compromised, then the credentials are exposed, potentially compromisingand can potentially used to compromise other systems and services. In addition, such credentials make it really difficult to periodically change the password. Even in AWS we sometimes see developers hard-code API Key IDs and Keys into apps in order to get access to some AWS service. This is a security accident waiting to happen, and can be avoided through the use of IAM Roles.

With AWS, we can assign a single IAM Role to an EC2 instance. This assignment is usually made when the instance is launched, but can also be done at runtime if needed. Applications running on the server retrieve the Role’s security credentials by pulling them out of the instance metadata through a simple web command. These credentials have an additional advantage over potentially long-lived, hard-coded credentials, in that they are changed or rotated frequently, so even if somehow compromised, they can only be used for a brief period.

Another key security advantage of Roles is that they can be limited to just the access/rights privileges needed to get a specific job done. Amazon’s documentation for roles gives the example of an application that only needs to be able to read files out of S3. In this case, one can assign a Role that contains read-only permissions for a specific S3 bucket, and the Role’s configuration can say that the role can only be used by EC2 instances. This is an example of the security principle of “least privilege,”, where the minimum privileges necessary are assigned, limiting the risk of damage if the credential is compromised. In the same sense that you would not give all of your users “Administrator” privileges, you should not create a single “Allow Everything” Role that you assign everywhere. Instead create a different Role specific to the needs of each system or group of systems.

Delegation

Sometimes one company needs to give access to their resources to another company. Before IAM Roles, (and before AWS) the common ways to do that were to share account logins (with the same issues identified earlier with hardcoded credentials) or to use complicated PKI/certificate based systems. If both companies using AWS, sharing access is much easier with Role-based Delegation. There are several ways to configure IAM Roles for delegation, but for now we will just focus on delegation between accounts from two different organizations.

At ParkMyCloud, our customers use Delegation to let us read the state of their EC2, RDS, and scaling group instances, and then start and stop them per the schedules they configure in our management console.

To configure Role Delegation, a customer first creates an account with the service provider, and is given the provider’s AWS Account ID and an External ID. The External ID is a unique number for each customer generated by the service provider.

The administrator of the customer environment creates an IAM Policy with a constrained set of access (principle of “least privilege” again), and then assigns that policy to a new Role (like “ParkMyCloudAccess”), specifically assigned to the provider’s Account ID and External ID.  When done, the resulting IAM Role is given a specific Amazon Resource Name (ARN), which is a unique string that identifies the role.  The customer then enters that role in the service provider’s management console, which is then able to assume the role.  Like the EC2 example, when the ParkMyCloud service needs to start a customer EC2 instance, it calls the AssumeRole API, which verifies our service is properly authenticated, and returns temporary security credentials needed to manage the customer environment.

Conclusions

AWS IAM Roles make some tasks a lot simpler by flexibly assigning roles to instances and other accounts. IAM Roles can help make your environment more secure by:

  • Using the principle of Least Privilege in IAM policies to isolate the systems and services to only those needed to do a specific job.
  • Prevent hard coding of credentials in code or files, minimizing danger from exposure, and removing the risk of long-unchanged passwords.
  • Minimizing common accounts and passwords by allowing controlled cross-account access.