LAS VEGAS — Most of us consider the scale of public clouds to be vast, but for a company the size of Thomson Reuters, it’s not nearly enough.
“The kick in the head for me is: ‘Yeah, our databases only go up to a terabyte.’ I throw up a terabyte database every day,” said Mark Bluhm, the company’s senior vice president of global technology operations.
Bluhm was one of a few end users speaking at the inaugural future:net conference yesterday, and his message was particularly relevant given the nature of the conference.
Created by VMware and hosting about 180 in this first go-around, future:net is meant to be a networking conference that takes into account the massive changes of the cloud, particularly when it comes to enterprise IT. As VMware chief technology strategy officer Guido Appenzeller put it, “You have to suddenly create networks where the hardware is no longer owned by you.”
So, future:net aims to more thoroughly bring cloud providers into the IT networking discussion. But Bluhm’s talk, delivered early in the two-day conference, showed that the cloud isn’t magic. It doesn’t yet suit all needs, not even scale.
Killing 500 Data Centers
“The thing that strikes me in every one of these is that I’m talking to people that build infrastructure, and I’m talking at an infrastructure level,” he said. “They never talk to me about: ‘What do you need to build this company at scale?'”
Those clouds have scale, but Thomson Reuters needs scale. Bluhm estimates the company has 400 TB of data change per day. “That’s not new data. That’s data that changes every day that’s got to get replicated across the world.”
Thomson Reuters’ problems with scale start with its sheer size. The Thomson and Reuters sides of the company both grew via acquisition, and then, of course, the two halves merged. That added up to 550 data centers — counting any location with customer-facing products or more than 50 devices — when Bluhm started; he’s winnowed that down to 200 and has the ultimate goal of around 18 data centers.
The company runs 16,000 network devices, many of them at customer locations, and it operates one of the largest Hadoop clusters outside of academia, he said.
Bluhm tried building his own cloud but found it to be a waste. “People were using OpenStack as a way to get fast provisioning, but they weren’t writing cloud applications,” he said. Moreover, he realized he couldn’t keep up with the features offered by public clouds. So, Thomson Reuters began using a combination of AWS and Azure, and it’s had briefings with Google Cloud Platform as well.
Other challenges are specific to Thomson Reuters’ businesses. Data has to be collected worldwide, and in some cases, must remain in a particular country. The company offers legal services that require storing archived legal codes, because lawyers sometimes need to look up what the law used to be. And the nature of its data, particularly in financial services, means security and regulatory compliance loom large.
Here are a few of the areas where Bluhm would like to see the public clouds improve.
Connectivity. Bluhm showed slides outlining how his networks connect to Azure but noted that the plan will have to change. For his needs, it has too many points of failure and not enough bandwidth — in the long term, it won’t scale.
Service-level agreements. Apparently, Bluhm doesn’t have any with his cloud providers. “I don’t care what the SLA is, but I have to know what it is so I can adhere to something,” he said. “I know how hard a problem that is for you guys, but: Figure it out.”
As-a-service economics. Bluhm would like to turn his 16,000 network devices into infrastructure-as-a-service (IaaS) offerings, if only because he’ll never be able to upgrade all those devices in a reasonable timeframe. But with that many devices spread across an immense number of points of presence (PoPs), Thomson Reuters doesn’t enjoy the usual economics of IaaS. The pricing comes out to be more like a managed service, Bluhm said.
Bluhm was on a roll, so he shared a few choice thoughts about his software developers as well. Regarding the trend of continuous integration and continuous deployment (CI/CD), he said, “For about half my development teams, that’s their version of ‘not needing to plan.'” And he noted that if a newly graduated developer “had a single security class, it’s amazing.”
Photo: Pre-conference bustle at future:net.