Image: Peter Dazeley via Getty Images

Teaching computers to tell time

Protocol Enterprise

Hello and welcome to Protocol Enterprise! Today: how a new startup plans to solve an age-old enterprise tech problem, cloud desktops might finally be an idea that sticks, and three data scientists walk into a bar: This AI can predict what happens next.

Spin up

Site reliability engineers (SREs) are the people who keep the cloud running, and tend to only get noticed for their efforts when something goes wrong. Perhaps that’s why 85% of SREs are looking at automation and AIOps services for help as their businesses scale, according to Dynatrace.

No one told you when to run

The people who know the most about computers consider it nothing short of a miracle they actually work at all. A new enterprise startup called Clockwork wants to fix an old, fundamental problem with computer networks — their failure to keep very accurate time.

Its mission is to bring nanosecond clock-synchronization accuracy into distributed systems to empower time-sensitive applications used in cryptocurrency and stock trading, mobile banking, online gaming, database design and other industries.

  • “If you look at cloud computing, the big idea in the first decade of the century was virtualization,” Clockwork co-founder and CEO Balaji Prabhakar said. “The next decade was all big data, crunching large amounts of data. What we believe is that this decade is going to move on time, timeliness, deadlines, real-time control: things to do with more time sensitivity.”
  • Latency Sensei is Clockwork’s new network latency sensor for cloud, hybrid and on-premises data-center environments.
  • Underpinned by the company’s clock synchronization technology, the software measures one-way network delays and helps customers detect bottlenecks, hiccups and underperforming virtual machines to optimize application performance.

Company executives say Latency Sensei’s sensor cuts through the “fog of virtualization” to give DevOps engineers visibility into their networks’ underlying infrastructure.

  • “When you go to the cloud and rent some virtual machines, you're in a bubble,” Prabhakar said.
  • “You don't know what hardware you're running on, you don't know whether two virtual machines are in the same rack of servers or are they across each other in the data center. And when you don't know, you can't tell how much time was spent in the network.”
  • Latency Sensei promises to solve this problem, determining as precisely as possible how long it takes for a packet or any piece of data to go from one node in a network to another node.
  • “For 50 years … in networking, we've never really managed to measure one-way delays accurately, because we don't have accurate clocks,” Prabhakar said, referring to unsynchronized network clocks. “All we've done historically is to say, ‘The time from here to there is half the time it takes to go from here to there and coming back.’”
  • That round trip time approach isn’t adequate to determine where a delay is happening, because it doesn’t distinguish between forward congestion and reverse congestion that can be picked up with accurate one-way times, he said.

Clockwork launched in 2018 to commercialize clock synchronization research conducted at Stanford University under the supervision of Prabhakar and VMware co-founder Mendel Rosenblum, who serves as Clockwork’s chief scientist.

  • Latency Sensei’s underlying time-synchronization technology comes from Clockwork’s first software product, Clock Sync, which synchronizes clocks in computers at extremely high levels of accuracy: single-digit nanosecond accuracy for hardware timestamps and within hundreds of nanoseconds for software timestamps, according to the company.
  • The Clock Sync software launched for private data centers in March 2019 and for public clouds a year later. It can scale to thousands or tens of thousands of nodes.
  • Accurate clock synchronization is an old problem, and the jittery nature of networks connecting the clocks in servers makes it hard to resolve, because the networks could add random delays to packets exchanged by the clocks, according to Prabhakar.
  • “The network is kind of the enemy in this equation,” he said. “The approach taken was to synchronize the switches in the network with the reference clock and use the network to convey time. This is a hardware-based way of doing it — expensive and hard to scale.”

Clockwork came up with a software approach that doesn’t require a network upgrade.

  • “The way we went about it was to not touch the network, but make it more of a signal processing, machine-learning-type approach, by just cleaning up the timestamps that we get from all the random delay noise added by the network,” Prabhakar explained.
  • “We treat the network like a black box and make clock synchronization an application service. This makes us scalable and accurate. Could this have been done 30 years ago or 25 years ago, probably not — not the way we've done it, because the ML technology that we needed wasn't around.”

“Having computers synchronized is a really powerful thing,” said Greg Papadopoulos, an NEA partner and the former chief technology officer at Sun Microsystems.

  • “There's just so much promise here. It is really something that you don't get to see that often … where you can take such a fundamental idea that can transform an area and, at the same time, you're not really sure what that means. It's really going to be the story of how we taught computers to keep time.”
— Donna Goodison (email | twitter)


DuckDuckGo has an all-in-one privacy solution aimed at simplifying online privacy protection. DuckDuckGo’s app can be used as an everyday browser with private search, tracker blocking, encryption, and now email protection built-in. It’s the free, easy button for online privacy.

Learn more

Armonk on the main line

For all the criticism visited on mainframes in the early days of the pandemic, which revealed just how many U.S. states rely on creaky, aging tech infrastructure for day-to-day operations, these machines continue to have remarkable staying power.

IBM unveiled the latest generation of its venerable mainframe product line Tuesday, called the z16. It runs on a custom IBM-designed processor and while most mainframe customers use them for transaction processing, the new machine can also handle AI workloads.

“Despite the ongoing move to the cloud, IBM says two-thirds of the Fortune 100, 45 of the world’s top 50 banks, eight of the top 10 insurers, seven of the top 10 global retailers and eight out of the top 10 telcos rely on its mainframes for critical processes,” according to Data Center Dynamics, underscoring the role mainframes continue to play in enterprise tech.

But IBM just couldn’t resist overplaying its hand, declaring that the new systems will be “the industry’s first quantum-safe system[s],” implying they’ll be able to withstand attacks from future quantum computers that have yet to be built. In the fine print below its press release, IBM attributed this claim to “a third party analyst” who was referring to NIST’s list of post-quantum cryptography algorithms, which is very much a work in progress.

And you’ll also need to upgrade to the Crypto Express 8S card to be “quantum safe.”

— Tom Krazit (email | twitter)

Upcoming at Protocol

It’s been almost six months since Congress passed the landmark $1 trillion Infrastructure Investment and Jobs Act. What progress toward those goals have we seen so far — and what can we expect in the next six months?

In this Protocol virtual event on April 21 at 9 a.m. PT, we will explore how the infrastructure bill rollout is going and what it means for you. Join Protocol’s Issie Lapowsky in conversation with Alan Davidson, assistant secretary for Communications and Information, U.S. Department of Commerce; Nicol Turner Lee, senior fellow and director of the Center for Technology Innovation, The Brookings Institution; and Angela Siefer, executive director, National Digital Inclusion Alliance. RSVP here.

What’s the deal with machine learning?

AI-powered language models are less than perfect, but researchers keep plugging away at machines that can replicate human communication skills. Humor is a very useful and entertaining communication skill, and Google researchers said this week they’re making progress on a version of its Pathways Language Model (PaLM) that can explain jokes.

“Remarkably, PaLM can even generate explicit explanations for scenarios that require a complex combination of multi-step logical inference, world knowledge, and deep language understanding. For example, it can provide high quality explanations for novel jokes not found on the web,” researchers wrote. There’s a lot of work ahead, but at least that's an improvement over previous language models, which were themselves a joke.

— Tom Krazit (email | twitter)

Around the enterprise

Microsoft will give the virtual desktop another go with the launch of Windows 11, which will allow administrators to direct business users toward Windows 365 Cloud PC instances running in a browser.

Initial worries that the Spring4Shell vulnerability was as bad as the Log4j vulnerabilityseem to have passed, but Microsoft is nevertheless seeing attempts to exploit that flaw in different parts of its cloud empire.


Tracking is a comprehensive problem — over 80% of websites, apps and emails contain third-party trackers. Because of that, people need a multi-pronged privacy solution. DuckDuckGo’s all-in-one privacy app can be used as an everyday browser with multiple features built-in, including private search, tracker blocking, encryption, and email protection.

Learn more

Thanks for reading — see you tomorrow!

Recent Issues