SIGCOMM

\documentclass[sigconf,screen]{acmart}

\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{booktabs}
\usepackage{algorithm,algpseudocode}
\usepackage{listings}
\usepackage{xcolor}

\lstset{
  basicstyle=\ttfamily\footnotesize,
  columns=fullflexible,
  breaklines=true,
}

\AtBeginDocument{%
  \providecommand\BibTeX{{%
    \normalfont B\kern-0.5em{\scshape i\kern-0.25em b}\kern-0.8em\TeX}}}

\acmConference[SIGCOMM '26]{ACM SIGCOMM 2026 Conference}{September 8--12, 2026}{City, Country}
\acmISBN{978-1-4503-XXXX-X/26/09}
\acmDOI{10.1145/XXXXXXX.XXXXXXX}
\setcopyright{acmlicensed}
\copyrightyear{2026}
\acmYear{2026}

\begin{document}

\title{Chronos: Microsecond-Precision Clock Synchronization\\
       for Wide-Area Datacenters}

\author{First Last}
\affiliation{\institution{University of Example}\country{Country}}
\email{[email protected]}

\author{Jane Doe}
\affiliation{\institution{Example Research Labs}\country{Country}}
\email{[email protected]}

\author{John Smith}
\affiliation{\institution{University of Example}\country{Country}}
\email{[email protected]}

\renewcommand{\shortauthors}{Last et al.}

\begin{abstract}
Datacenter applications increasingly rely on high-precision time, but
existing protocols either lack accuracy at wide-area scale (NTP) or
require expensive hardware support end-to-end (PTP with boundary
clocks). Chronos is a clock synchronization system combining in-network
timestamping on programmable switches, Bayesian path-asymmetry
estimation, and an epoch-based synchronization protocol. Chronos
achieves microsecond-scale synchronization across geographically
distributed sites on commodity hardware, with p99 error of 9.8
$\mu$s across three transcontinental datacenters---a 45$\times$
improvement over NTP and 32$\times$ improvement over software PTP.
\end{abstract}

\begin{CCSXML}
<ccs2012>
<concept><concept_id>10003033.10003034.10003037</concept_id>
<concept_desc>Networks~Data center networks</concept_desc>
<concept_significance>500</concept_significance></concept>
</ccs2012>
\end{CCSXML}
\ccsdesc[500]{Networks~Data center networks}

\keywords{clock synchronization, datacenter networks, PTP, programmable switches}
\maketitle

\section{Introduction}
Precision time enables lock-free concurrency, distributed tracing, and
commit ordering. Wide-area deployments expose NTP's millisecond limits
and software PTP's fragility. Hardware PTP achieves nanoseconds but
requires boundary-clock support end-to-end---rarely available across
WAN carrier links.

\paragraph{Contributions.}
\begin{itemize}
\item An in-network timestamping primitive on programmable switches.
\item A Bayesian estimator for path asymmetry that works without
  a priori link calibration.
\item An evaluation across three transcontinental datacenters showing
  consistent sub-10-$\mu$s p99 error.
\end{itemize}

\section{Background}
We review NTP, PTP, TrueTime, and recent programmable-switch work. The
key challenge for wide-area PTP is \emph{path asymmetry}: the forward
and reverse paths traverse different routers and often different links,
breaking PTP's symmetric-delay assumption.

\section{Design}
Chronos couples three ideas: (a) in-network timestamping using
programmable switches, (b) Bayesian path-asymmetry estimation, and
(c) an epoch-based synchronization protocol that is resilient to
packet loss.

\subsection{Path-Asymmetry Model}
For a four-message PTP-style exchange with timestamps $t_1, t_2, t_3, t_4$:
\begin{equation}
\hat\Delta = \frac{(t_4 - t_1) - (t_3 - t_2)}{2} - \alpha_\text{asym}.
\end{equation}
We estimate $\alpha_\text{asym}$ online from history using a hierarchical
Bayesian model with priors drawn from traceroute-level topology.

\subsection{Epoch Protocol}
\begin{algorithm}[t]
\caption{Chronos epoch synchronization (simplified)}
\begin{algorithmic}[1]
\State upon epoch boundary
\State exchange $N$ timestamped messages with reference
\State compute $\hat\Delta$ using the Bayesian estimator
\State gradually adjust local clock by $\hat\Delta$ over the next epoch
\end{algorithmic}
\end{algorithm}

\section{Implementation}
The data plane runs on P4 Tofino switches. The control plane is 6{,}300
lines of Rust. Chronos exports a POSIX \texttt{clock\_gettime}-compatible
interface.

\section{Evaluation}
\subsection{Methodology}
We deploy Chronos across three datacenters on two continents. Reference
time is a GPS-disciplined oscillator at each site. Measurement workload
mimics Spanner-style commit ordering.

\begin{table}[t]
\centering
\begin{tabular}{lrr}
\toprule
Protocol & p50 error ($\mu$s) & p99 error ($\mu$s) \\
\midrule
NTP (stratum-1)  &  980 & 4{,}600 \\
Software PTP     &   45 &    320 \\
\textbf{Chronos} & \textbf{1.2} & \textbf{9.8} \\
\bottomrule
\end{tabular}
\caption{Synchronization error across three wide-area datacenters.}
\label{tab:main}
\end{table}

\subsection{Analysis}
Bayesian asymmetry estimation accounts for the majority of the gap over
software PTP. In-network timestamping contributes an additional
$\sim$3$\times$ by eliminating kernel-timestamping jitter.

\section{Related Work}
Huygens, Sundial, Graham; prior PTP deployments.

\section{Conclusion}
Microsecond clock synchronization across wide-area datacenters is
tractable when we combine programmable data-plane support with
principled asymmetry estimation.

\section*{Ethics Considerations}
This work does not involve human subjects and raises no direct ethical
concerns.

\begin{acks}
We thank our SIGCOMM shepherd and the anonymous reviewers.
\end{acks}

\bibliographystyle{ACM-Reference-Format}
\bibliography{refs}

\end{document}