Twitter CEO Jack Dorsey: Improved Uptime Is Top Priority

Twitter is working on adding the ability for users to separate their contacts into groups, support for sending and receiving messages by e-mail, and adding new instant-message and Internet channels for the service, said CEO Jack Dorsey in a one-on-one interview with <i>InformationWeek.</i> But first Twitter needs to fix the outages that have plagued the service for weeks -- that's the first and only priority for now, he said. We talked with Dorsey about uptime, the company's business model, and

Mitch Wagner, California Bureau Chief, Light Reading

June 4, 2008

4 Min Read

Twitter is working on adding the ability for users to separate their contacts into groups, support for sending and receiving messages by e-mail, and adding new instant-message and Internet channels for the service, said CEO Jack Dorsey in a one-on-one interview with InformationWeek. But first Twitter needs to fix the outages that have plagued the service for weeks -- that's the first and only priority for now, he said. We talked with Dorsey about uptime, the company's business model, and new features."We've been adding some databases and hardware and tweaking our schema to take a lot of load off the main database," Dorsey said late Tuesday. "We've seen some improvements in the last few days, and we're making sure we make a good record of stability before we turn on IM, which is a popular service and will increase the load." Twitter users can send and receive messages -- called "tweets" -- using a variety of channels, including the Web, text messaging, special-purpose applications such as Twitterific and Twhirl, RSS, and instant messaging; however, IM support for Twitter was turned off during the past couple of weeks when the service kept crashing. As of the last report, IM was still off, as was a service called "tracking," through SMS, where users can set a keyword to watch and any tweet from any user containing that keyword will be sent to them.

The outage was caused by increased demand on the service, which was not designed to take the load. The service runs on MySQL on Red Hat Enterprise Linux, on a managed hosting service NTT Managed Hosting Platform.

I was surprised to hear that Twitter is running on a managed hosting service; most mature dot-coms run their own data center. Dorsey agreed. "It's a transition that everyone makes when the time is right," Dorsey said. The Twitter team is constantly considering whether to build its own data center. However, by running in a managed hosting service, Twitter is hiring people who can handle system maintenance, so that's a trade off. Twitter actually started out running its own servers in a co-location facility, but decided to move to managed hosting because of the immediate benefits. "It's a decision we're coming up against again and again," he said.

Twitter recently hired engineer Steve Jensen and distributed systems architect John Kalucki to help improve the infrastructure. Jensen worked on Blogger at Google, and Kalucki founded middleware vendor SQLstream and moved on to DoubleClick, where he helped scale their infrastructure. "They've been with the team for over two months now, and we continue to look for amazing engineers and operations people. We're still hiring," Dorsey said. Google acquired DoubleClick in March. Blogger was founded by Twitter co-founder Ev Williams, and was acquired by Google in 2003.

I asked Dorsey whether extremely popular users create a burden on the system. The Twitter blog said recently that popular users contribute to downtime, which offended überblogger Robert Scoble, who is following 21,000 people on Twitter, and is followed by 26,000, making him the fifth most popular user on Twitter. By comparison, Williams is #18 and Dorsey is #67. (Borat is #95.)

"Any massive user is going to make the system work, certainly," Dorsey said. "The way we originally designed the architecture, and the way the system is designed right now, we have various pain points." Popular Twitter users create a chain reaction -- when they update, their followers are more likely to update, too, leading to usage surges. "When you get an update from a friend, you are likely to update as well," he said. "We have to design the system for that kind of flash-mob behavior."

Twitter is working on decoupling its architecture, transforming it into a service-based technology where the breakdown of a single service won't crash the entire system. For example, they might move the "Everyone" feed -- a stream of updates from everyone, around the world -- to a separate service, which could then be cached so that every hit on the Everyone feed doesn't require a database query and calculation.

This is the first part of my Dorsey interview. We also talked about upcoming service upgrades and Twitter's lack of a business model.

Do you have any questions for Jack? Leave a message below and I'll try get him to answer them.

Follow InformationWeek on Twitter.

About the Author(s)

Mitch Wagner

California Bureau Chief, Light Reading

Mitch Wagner is California bureau chief for Light Reading.

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like


More Insights