As public cloud service-level agreements come under fire from IT experts, HP execs say they plan to refine the...
company's Infrastructure as a Service uptime contracts for enterprise customers.
Following some confusion about the terms of Hewlett Packard Co.'s Cloud Compute service-level agreement last month, HP said it will add a preamble to the SLA that summarizes its overall intent and explains an example scenario, said Blake Yeager, product manager for HP Cloud Services.
There is an onus on the user today to detect an instance failure in HP Cloud Compute because instances are ephemeral, Yeager said. Another service that roots more persistent server instances in HP's block storage will be proactively monitored by HP in the future.
Frankly, many providers who have good SLAs still force you through a tangle of verbiage to figure out what they intend.
Lydia Leong, Gartner analyst
HP will also launch additional services for monitoring and notification of instance failures; this service will be in private beta mode next month. There's also an attempt afoot to improve the accuracy of the reporting that HP receives internally so it can send preemptive notifications to customers sometime mid-year.
Furthermore, HP will offer SLAs based on more than availability as its services mature, Yeager said.
"Where we plan on taking this as we progress is SLAs around the performance of the service, and then the availability of specific individual resources, and then the performance of those individual resources," he said.
In the event of a mass outage situation, HP has throttling in place at the application program interface (API) to prevent an overload of the control plane, Yeager added.
Meanwhile, Amazon Web Services (AWS) executives have repeatedly said reliability on its low-margin, high-volume infrastructure is a matter for customers to manage -- and some do just that.
High-profile AWS customer Netflix has open-sourced tools meant to improve the availability of applications running on Amazon's infrastructure.
HP's SLA confusion
These nuances of cloud SLAs confuse customers and analysts, who want vendors to simplify SLA terms.
This publication and Gartner analyst Lydia Leong both reported last month that HP's SLA required replication of workloads between all availability zones in a region, similar to the SLA offered by AWS.
HP clarified that it does not require users of its service to replicate data between all availability zones within a region. An HP spokesperson further stated that HP's SLA differs from Amazon's in that it is calculated monthly instead of yearly, resulting in more guaranteed uptime than is offered by AWS.
Still, Leong found fault with some aspects of HP's SLA in an updated blog post, chiefly that the SLA's language is unclear enough to allow for such misunderstandings.
"Both HP and AWS sin in this regard, and frankly, many providers who have good SLAs still force you through a tangle of verbiage to figure out what they intend," Leong wrote.
Other analysts go further in their criticisms of public cloud SLAs offered by both HP and Amazon, as well as other public cloud Infrastructure as a Service (IaaS) providers, and say these SLAs put too much of the onus for detecting and resolving failures on the end user rather than the service provider.
"The entire public cloud computing market is in disarray on this point," said Bernd Harzog, an analyst with The Virtualization Practice.
For example, HP's SLA requires the user to detect that an instance has failed, and begins to count downtime in the initial six minute period during which a customer makes a failed call to the HP API. Leong's updated post was critical of this, as in her eyes, users attempting to restart an instance could potentially compromise the performance of the API in an outage situation.
Simple up-or-down availability is a poor measure of IaaS quality of service, Harzog added.
"If you're trying to do an online financial transaction and it takes 10 seconds, is that acceptable?" he said. "Right now there's no ability on the part of any public cloud vendor to guarantee how long something takes."
As AWS expands its enterprise cloud base, it introduces a catch-22 for itself as well as other IaaS vendors, analysts say. The AWS high-volume, low-margin infrastructure competes with established IT vendors' fledgling services on cost, often at the expense of availability.
Still, IT vendors who attempt to compete with Amazon by offering enhanced availability are still often asked why they are more expensive than AWS, experts said.
Meanwhile, "if Amazon really wants to get into the enterprise, they're going to have to make some changes to their SLAs," said Larry Carvalho, a cloud computing consultant with Robust Cloud LLC. "But I'm not sure they're ready for that."