Milliman, one of the world's largest actuarial firms, extols the benefits of SaaS to its insurance users.
Cloud Contracts: 8 Questions To Ask
(Click image for larger view and slideshow.)
When it comes to the cloud, respondents to an InformationWeek survey were emphatic about the value of software-as-a-service (SaaS): 55% believe software service in the cloud delivers better results than on-premises systems. Actuarial firm Milliman would agree.
Milliman is in the process of transitioning a major part of its business from licensed software on premises to SaaS because its customers -- actuarial experts at insurance companies -- are finding that the spikes in their workloads match up nicely with quickly assembled clusters in the cloud, followed by their dismantling at the end of the quarter.
Brian Reid, global sales director for Milliman's MG-Asset Liability Financial Analysis actuarial system, is leading the transition of MG-ALFA to a SaaS offering. Milliman offers a Windows version as an on-premises system that can define risk and price insurance coverage by running thousands of simulations that represent conditions likely to be encountered by an insurance package over a 30-year period. The simulations are compute-intensive, and in the past were frequently run on a special-purpose cluster of a thousand servers or more.
The Phoenix Group, one of Milliman's customers in the UK, wanted to do the modeling and meet increased reporting needed by Solvency II requirements that take effect soon in Europe -- and it wanted to do so without the capital expenditure of a another large actuarial cluster. Some firms use up to a thousand servers in such clusers, said Reid. Such a cluster would be needed at the end of each quarter, but most of the rest of the time it would sit idle or underused. Milliman revamped MG-ALFA for the Microsoft Azure cloud, where the Phoenix Group could activate it on command on the number of servers that it wanted.
"Phoenix realized that the workloads to meet the new regulations would be significantly higher than anything they were currently doing, but they would be intermittent," Reid said in an interview. "The new requirements dictated use of a large cluster at the end of each quarter but it would be idle much of the rest of the time. That meant a capital expense, with the investment sitting underutilized."
MG-ALFA on Azure worked for Phoenix, and Milliman has since made the system available on Azure on a general-purpose basis to its other customers (the first version was specific to Phoenix). Milliman had to write a scheduler for MG-ALFA to function as a multi-tenant system and create the right pipelines for it to communicate with other resources on Azure. But Reid said the investment has proved worth it by opening discussions with new customers. Originally installed in one Azure center, Milliman now has MG-ALFA available on up to 3,000 VMs in Dublin, 5,000 in the UK, and 2,000 in Amsterdam. With its new reporting requirements, Europe is a competitive battlefield for Milliman.
Although there are competing actuarial systems on the market, Reid said, "we are finding it to be a distinct competitive advantage." Customers offered a large block of business can now turn to Milliman's system on Azure, analyze it, and decide how to price it without making a large capital investment in compute power.
For those who didn't wish to refresh an aging in-house cluster or wanted to avoid the capital cost of building a cluster, Reid added, "using 1,000 machines for 10 hours costs the same as 10,000 for one hour."
Two years ago, Reid said, he saw more customer skepticism about security and reliability in the cloud. Over the last 18 months, it's "completely changed," replaced by a view of the cloud as a competitive advantage, he said, partly because Milliman has figured in the cost of those actuarial clusters they've been using in-house. "The tendency is for them to sit idle 2.5 months of every quarter. If they were [used], then the cost savings of moving to the cloud become enormous," he said. If they're used 60% of the time -- an optimistic figure -- the Azure system would still save an insurance company $1 million of a $3 million.
Over the last 12 months, use of the SaaS version on Azure has increased 600%, according to Reid, although he declined to say exactly how many customers that represents.
A technical white paper on the Milliman site claims that in production environments over an 18-month period, the MG-ALFA system has clocked 30 million compute hours on grids from 50 to 10,000 cores with a greater-than-99.99% job completion rate.
In the InformationWeek survey, one respondent complained about the difficulty of comparing the cost of this type of system in the cloud to one on the premises. What's needed, wrote Sandy Barsky, IT specialist at the General Services Administration, is a cost translator between how software is used on premises versus how it's used in the cloud: "Restate ROI as return on use. Cloud is about use… There needs to be a tool to compare the software license offered for on-premises versus the same software as a service."
Milliman is one of the few places where such a tool is being used, built with the help of Microsoft, said Reid. It's instrumental in helping the firm convince customers they can cost-justify moving from the on-premises version to the SaaS version.
Charles Babcock is an editor-at-large for InformationWeek and author of Management Strategies for the Cloud Revolution, a McGraw-Hill book. He is the former editor-in-chief of Digital News, former software editor of Computerworld and former technology editor of Interactive ... View Full Bio
SaaS As Innovation Driver?Software as a service is the clear No. 1 way enterprises consume cloud. InformationWeek's SaaS Innovation Survey reveals three tips to get the most from SaaS: Make it a popularity contest. Have an escape plan. And remember that identity is the new perimeter.
InformationWeek Tech Digest August 03, 2015The networking industry agrees that software-defined networking is the way of the future. So where are all the deployments? We take a look at where SDN is being deployed and what's getting in the way of deployments.