We’re reaching the end of year 2018 and it is time to revisit cloud pricing.
The current quarter marks the historic acquisition of RedHat by IBM for $33B and it is a good opportunity to focus the comparison with IBM SoftLayer in mind.
A quick timeline of cloud pricing in case you didn’t follow:
- 2016: Google Cloud Compute reaches an equivalent breadth of infrastructure services with AWS, while instances are 50% cheaper in average.
- 2017: Google Cloud releases high memory instances (384GB and above), pricing on-par with IBM SoftLayer.
- 2018: AWS refreshes their range of (high memory) instances and pricing. Continue reading below.
Quick reminder: IBM SoftLayer is used to order beefy dedicated servers, including but not limited to terabytes of memory and a dozen of local SSDs. You can build a server by selecting individual components (enclosure, CPU, RAM). It is similar to ordering hardware from Dell except it’s leased to IBM and it stays in their data center.
The only hedge of SoftLayer is really exclusively to provide large servers that other competitors didn’t offer. It is a weak differentiator and it was only a matter of time before competitors would broaden their range. Refer to the previous article for more details, The Inevitable Demise of IBM SoftLayer by AWS.
AWS and Google released high memory instances and the time to catch up with SoftLayer is now.
We’ll look at high memory instances from the major cloud providers and equivalent servers from SoftLayer.
The typical use case for this type of hardware is databases, business intelligence solutions, big data and the intersection of these three. It’s all memory bound and memory is the single decisive factor when selecting hardware. Charts are sorted by memory to reflect this.
Price may vary slightly per region. AWS and Google instances are from the Oregon region, it had the latest offerings.
SoftLayer servers are from Washington. The selected server always includes a pair of disks for the OS (RAID 1), redundant power supplies and redundant 10 Gbps NIC.
Disks are not included (besides an OS for SoftLayer). A large number of disks may be attached to store data, including SSD, HDD, local and network storage. There is a range of storage options available at different price/performance points, from each of the provider, it is not covered in this article.
Monthly costs as given by the provider.
For Google, it includes the sustained used discount, a discount automatically applied when instances stay active for more than 2 weeks, up to 30% off for a full month.
Kindly note that Google allows to attach any number of local SSDs to any type of instance, unlike AWS that has a duplicated lineup with fixed local disks (r5 vs r5d).
Monthly costs with a yearly engagement.
For AWS, this is a reserved instance 1 year full upfront (30 to 50% off).
Despite doing this comparison, I generally advise against reserving any capacity. It consumes a lot of cash upfront and lot of time to analyze usage, to the point that the savings are almost always less than the investment.
AWS reservations have stringent restrictions attached and often can’t be changed even though hinting otherwise. Google reservations are barely better than the automatic monthly discount they substitute to and not really worth looking into.
In infrastructure, the most expensive cost is not to buy hardware upfront but to buy some hardware and have to buy some more later as needs invariably change. The cloud saves endless money by having the flexibility to add and remove resources as projects evolve, grow or die. Adjustments should happen over weeks or months and cannot be done at all if engaged.
Cloud providers are now on-par with SoftLayer in terms of hardware offering and pricing. This will intensify as AWS/Google continue to improve their services while IBM does not.
Considering that IBM is worse in every single aspect including but not limited to API availability and documentation, network and DNS services, storage services, etc… there is pretty much zero reason whatsoever to use IBM nowadays.
The third cloud in the run is Microsoft Azure, that was omitted in this article. I shall start writing about it at some point.
To conclude this:
- Go to AWS if you want the industry standard, noone ever got fired for choosing AWS.
- Go to Google Cloud if you want the same as AWS, at a better price and with better performances. Especially applicable to tech companies and future unicorns.
- Go to Microsoft Azure if you want… just kidding you didn’t want Azure but their sales force convinced your boss otherwise. Especially applicable to Fortune 100.
- Realistically, any large business should have at least 2 of those for resiliency.