The mainstream media’s coverage and interest in cloud computing has been an exciting boon to our industry, although I’m not sure if I should laugh or cry every time I see an article mentioning an outage of a notable public cloud. I suppose it’s inevitable that with increased attention comes increased scrutiny and as they say in the newspaper business “if it bleeds it leads” – that seems to be the approach many journalists have taken to covering cloud outages.
The reality is networks crash. Routers and hard drives fail. While all data center operators and hosting/cloud providers take measures to build in various layers or redundancy, outages happen. This has and always will be the case for hosted services. In the end, this is why service level agreements (SLAs) exist and what truly matters is a provider’s transparency, customer service and credits.
To this end, I thought it would be an interesting exercise to compare the two leading infrastructure cloud services SLAs from Rackspace and Amazon. Although this meant coping with flashbacks from the 84-page report I wrote for Tier 1 Research back in 2001 on this very topic comparing the SLAs of the top 20 MSPs of the era, and I found it to be rather illuminating. Here’s a quick reference table:
| Rackspace – Cloud Servers | Amazon – EC2 | |||||||||||||||||||
| Uptime / Availability Guarantee | 100% | 99.95% | ||||||||||||||||||
| Time span | Current period | “service year” or the preceding 365 days | ||||||||||||||||||
| Time-to-resolution | 1 hour | Not specified | ||||||||||||||||||
| Credits | · 5% of the fees for each 30 minutes of network or data center downtime, up to 100% of the fees· 5% of the fees for each additional hour of downtime past time-to-resolve, up to 100% of the fees | 10% of bill per eligible credit period | ||||||||||||||||||
| Notification onus | Customer | Customer | ||||||||||||||||||
| Window | 30 days after incident | 30 days after incident | ||||||||||||||||||
| Rackspace – Cloud Files | Amazon – S3 | |||||||||||||||||||
| Availability | 99.9% | 99.9% | ||||||||||||||||||
| Definition | (i) The Rackspace Cloud network is down, or (ii) the Cloud Files service returns a server error response to a valid user request during two or more consecutive 90 second intervals, or (iii) the Content Delivery Network fails to deliver an average download time for a 1-byte reference document of 0.3 seconds or less, as measured by The Rackspace Cloud’s third party measuring service. | “Error Rate” means: (i) the total number of internal server errors returned by Amazon S3 as error status “InternalError” or “ServiceUnavailable” divided by (ii) the total number of requests during that five minute period. We will calculate the Error Rate for each Amazon S3 account as a percentage for each five minute period in the monthly billing cycle. | ||||||||||||||||||
| Credits |
|
|
Key Takeaways:
· It was quite interesting to see the different approaches taken by the two. On the Cloud Servers side, Rackspace service guarantee of 100% uptime is a long-standing marketer’s tactic, which simply means they will pay for any down time that does occur. Amazon on the other hand has a more realistic guarantee of 99.95%, which actually translates into just over 4.3 hours of non-scheduled downtime a year.
· The fact that Rackspace specifies a time-to-resolve guarantee and offers credits if it misses speaks to its heritage as a managed hoster first. This is something Amazon can and should put into place.
· In terms of credits, Amazon caps their credit at 10% per period compared to Rackspace which will ultimately provide 100% credit if it is warranted. Separate but related on Storage, Amazon caps the credit at 25% whereas again Rackspace offers up to 100% if warranted. Again, Rackspace’s approach is much more customer-centric.
· Both Rackspace and Amazon put the burden of an SLA violation notification and credit request on their customer. This would be my biggest critique of both firms’ SLAs. It is something I advocated back in 2001 and nothing has changed: customers are already frustrated by an outage – why make them bear the administrative responsibility to prove the outage existed and chase you down for the credit owed? Other cloud providers may want to consider offering an automated credit function when outages occur (think Orbitz.com if someone books the same room as you at a lower price) as a way to differentiate from the market leaders.
In conclusion, while auto-paying for SLA violations would be an improvement for both Amazon and Rackspace, I also recommend all cloud service providers post their uptime front-and-center on their website. At the end of the day, the credit issued for downtime (even if it was automatic) tends to pale in comparison to the acute frustration and anxiety the customer is experiencing. Ultimately, the best SLA would be “our CEO will fly to your office personally to explain what happened” but we all know that just isn’t scalable.
Source Links
http://www.rackspacecloud.com/legal/sla
http://aws.amazon.com/ec2-sla/
About Joshua Beil
Joshua Beil (@joshbeil) is the Director of Market Strategy and Research for Parallels, the leading provider of automation and virtualization software to the web hosting community. Previously, Josh was the Director of Social Media & Technology at a Level 3 Communications [Nasdaq: LVLT] where he provided strategic and tactical sales support. Before that, Josh was CEO and cofounder of Skywave Broadband, Inc, the largest commercial WiFi service provider in Hawaii. He was named one of Pacific Business News' Forty Under 40 for 2006, and in 2005, he was named a High Tech Leader by the Pacific Technology Foundation. Before co-founding Skywave, Mr. Beil was VP of Research and Development for the market research boutique, Tier 1 Research, where he covered the Internet infrastructure sector as an analyst, and negotiated and sold subscriptions to Tier 1's research services. Josh has also previously served as the Senior Analyst for Exodus Communications as well as the internationally known market research firm, IDC. He holds a Certificate in E-Business from UCSC Extension, and he graduated with honors from the University of California at Santa Cruz with a major in Psychology.
No related posts.
{ 1 comment… read it below or add one }
Excellent article. I believe the relationship between provider and customer will change soon. Transparent SLA must be provided to the customer and the payback should be automatic. We would never buy any goods without a minimum guarantee. Why this should be different for cloud servers or cloud services ?