CINXE.COM

High Resolution Time

<!DOCTYPE html> <html lang="en"> <head> <meta charset="utf-8"> <title> High Resolution Time </title> <script src="https://www.w3.org/Tools/respec/respec-w3c" async class= "remove"></script> <script class="remove"> var respecConfig = { shortName: "hr-time-3", specStatus: "ED", editors: [ { company: "Google LLC", companyURL: "https://google.com/", name: "Yoav Weiss", url: "https://blog.yoav.ws/", w3cid: "58673", }, ], formerEditors: [ { company: "Google LLC", name: "Ilya Grigorik", retiredDate: "2021-03-01", url: "mailto:igrigorik@gmail.com", w3cid: "56102", }, { company: "Google LLC", name: "James Simonsen", retiredDate: "2015-01-01", url: "mailto:simonjam@google.com", w3cid: "45726", }, { company: "Microsoft Corp.", name: "Jatinder Mann", retiredDate: "2014-02-01", url: "mailto:jmann@microsoft.com", w3cid: "44357", }, ], github: "w3c/hr-time", caniuse: "high-resolution-time", testSuiteURI: "https://wpt.live/hr-time/", implementationReportURI: "https://wpt.fyi/hr-time/", group: "webperf", xref: "web-platform", mdn: "hr-time", highlightVars: true, localBiblio: { "Temporal": { href: "https://tc39.es/proposal-temporal/", title: "Temporal", status: "Stage 3 Proposal", publisher: "ECMA TC39", }, }, }; </script> </head> <body> <section id="abstract"> <p> This specification defines an API that provides the time origin, and current time in sub-millisecond resolution, such that it is not subject to system clock skew or adjustments. </p> </section> <section id="sotd"></section> <section id="introduction" class="informative"> <h2> Introduction </h2> <p> The ECMAScript Language specification [[ECMA-262]] defines the <code><dfn data-no-export="" data-cite= "ECMA-262#sec-date-objects">Date</dfn></code> object as a time value representing time in milliseconds since 01 January, 1970 UTC. For most purposes, this definition of time is sufficient as these values represent time to millisecond precision for any moment that is within approximately 285,616 years from 01 January, 1970 UTC. </p> <p> In practice, these definitions of time are subject to both clock skew and adjustment of the system clock. The value of time may not always be monotonically increasing and subsequent values may either decrease or remain the same. </p> <p> For example, the following script may record a positive number, negative number, or zero for computed <code>duration</code>: </p> <pre class="example js"> var mark_start = Date.now(); doTask(); // Some task var duration = Date.now() - mark_start; </pre> <p> For certain tasks this definition of time may not be sufficient as it: </p> <ul> <li>Does not have a stable monotonic clock, and as a result, it is subject to system clock skew. </li> <li>Does not provide sub-millisecond time resolution. </li> </ul> <p> This specification does not propose changing the behavior of <code><dfn data-no-export="" data-cite= "ecma-262#sec-date.now">Date.now()</dfn></code> [[ECMA-262]] as it is genuinely useful in determining the current value of the calendar time and has a long history of usage. The {{DOMHighResTimeStamp}} type, {{Performance}}.{{Performance/now()}} method, and {{Performance}}.{{Performance/timeOrigin}} attributes of the {{Performance}} interface resolve the above issues by providing monotonically increasing time values with sub-millisecond resolution. </p> <p class="note"> Providing sub-millisecond resolution is not a mandatory part of this specification. Implementations may choose to limit the timer resolution they expose for privacy and security reasons, and not expose sub-millisecond timers. Use-cases that rely on sub-millisecond resolution may not be satisfied when that happens. </p> <section id="use-cases" class="informative"> <h3> Use-cases </h3> <p> This specification defines a few different capabilities: it provides timestamps based on a stable, monotonic clock, comparable across contexts, with potential sub-millisecond resolution. </p> <p> The need for a stable monotonic clock when talking about performance measurements stems from the fact that unrelated clock skew can distort measurements and render them useless. For example, when attempting to accurately measure the elapsed time of navigating to a Document, fetching of resources or execution of script, a monotonically increasing clock with sub-millisecond resolution is desired. </p> <p> Comparing timestamps between contexts is essential e.g. when synchronizing work between a {{Worker}} and the main thread or when instrumenting such work in order to create a unified view of the event timeline. </p> <p> Finally, the need for sub-millisecond timers revolves around the following use-cases: </p> <ul> <li>Ability to schedule work in sub-millisecond intervals. That is particularly important on the main thread, where work can interfere with frame rendering which needs to happen in short and regular intervals, to avoid user-visible jank. </li> <li>When calculating the frame rate of a script-based animation, developers will need sub-millisecond resolution in order to determine if an animation is drawing at 60 FPS. Without sub-millisecond resolution, a developer can only determine if an animation is drawing at 58.8 FPS (1000ms / 16) or 62.5 FPS (1000ms / 17). </li> <li>When collecting in-the-wild measurements of JS code (e.g. using User-Timing), developers may be interested in gathering sub-milliseconds timing of their functions, to catch regressions early. </li> <li>When attempting to cue audio to a specific point in an animation or ensure that the audio and animation are perfectly synchronized, developers will need to accurately measure the amount of time elapsed. </li> </ul> </section> <section id="examples" class="informative"> <h3> Examples </h3> <p> A developer may wish to construct a timeline of their entire application, including events from {{Worker}} or {{SharedWorker}}, which have different [=environment settings object/time origins=]. To display such events on the same timeline, the application can translate the {{DOMHighResTimeStamp}}s with the help of the {{Performance}}.{{Performance/timeOrigin}} attribute. </p> <pre class="example js"> // ---- worker.js ----------------------------- // Shared worker script onconnect = function(e) { var port = e.ports[0]; port.onmessage = function(e) { // Time execution in worker var task_start = performance.now(); result = runSomeWorkerTask(); var task_end = performance.now(); } // Send results and epoch-relative timestamps to another context port.postMessage({ 'task': 'Some worker task', 'start_time': task_start + performance.timeOrigin, 'end_time': task_end + performance.timeOrigin, 'result': result }); } // ---- application.js ------------------------ // Timing tasks in the document var task_start = performance.now(); runSomeApplicationTask(); var task_end = performance.now(); // developer provided method to upload runtime performance data reportEventToAnalytics({ 'task': 'Some document task', 'start_time': task_start, 'duration': task_end - task_start }); // Translating worker timestamps into document's time origin var worker = new SharedWorker('worker.js'); worker.port.onmessage = function (event) { var msg = event.data; // translate epoch-relative timestamps into document's time origin msg.start_time = msg.start_time - performance.timeOrigin; msg.end_time = msg.end_time - performance.timeOrigin; reportEventToAnalytics(msg); } </pre> </section> </section> <section id="sec-concepts"> <h3> Time Concepts </h3> <section id="sec-clocks"> <h4> Clocks </h4> <p> A <dfn>clock</dfn> tracks the passage of time and can report the <dfn>unsafe current time</dfn> that an algorithm step is executing. There are many kinds of clocks. All clocks on the web platform attempt to count 1 millisecond of clock time per 1 millisecond of real-world time, but they differ in how they handle cases where they can't be exactly correct. </p> <ul> <li>The <dfn data-export="">wall clock</dfn>'s <dfn data-dfn-for= "wall clock" data-export="" id= "wall-clock-unsafe-current-time">unsafe current time</dfn> is always as close as possible to a user's notion of time. Since a computer sometimes runs slow or fast or loses track of time, its [=wall clock=] sometimes needs to be adjusted, which means the [=wall clock/unsafe current time=] can decrease, making it unreliable for performance measurement or recording the orders of events. The web platform shares a [=wall clock=] with [[ECMA-262]] <dfn data-no-export="" data-cite= "ECMA-262#sec-time-values-and-time-range">time</dfn>. </li> <li> <p> The <dfn data-export="">monotonic clock</dfn>'s <dfn data-export= "" data-dfn-for="monotonic clock" id= "monotonic-clock-unsafe-current-time">unsafe current time</dfn> never decreases, so it can't be changed by system clock adjustments. The [=monotonic clock=] only exists within a single execution of the [=user agent=], so it can't be used to compare events that might happen in different executions. </p> <p class="note"> Since the [=monotonic clock=] can't be adjusted to match the user's notion of time, it should be used for measurement, rather than user-visible times. For any time communication with the user, use the wall clock. </p> <p class="note"> The user agent can pick a new [=estimated monotonic time of the Unix epoch=] when the browser restarts, when it starts an isolated browsing session—e.g. incognito or a similar browsing mode—or when it creates an [=environment settings object=] that can't communicate with any existing settings objects. As a result, developers should not use shared timestamps as absolute time that holds its monotonic properties across all past, present, and future contexts; in practice, the monotonic properties only apply for contexts that can reach each other by exchanging messages via one of the provided messaging mechanisms - e.g. {{Window/postMessage(message, options)}}, {{BroadcastChannel}}, etc. </p> <p class="note"> In certain scenarios (e.g. when a tab is backgrounded), the user agent may choose to throttle timers and periodic callbacks run in that context or even freeze them entirely. Any such throttling should not affect the resolution or accuracy of the time returned by the monotonic clock. </p> </li> </ul> </section> <section> <h4> Moments and Durations </h4> <p> Each [=clock=]'s [=unsafe current time=] returns an <dfn>unsafe moment</dfn>. [=Coarsen time=] converts these [=unsafe moments=] to <dfn data-export="" data-lt="moment">coarsened moments</dfn> or just [=moments=]. [=Unsafe moments=] and [=moments=] from different clocks are not comparable. </p> <p class="note"> [=Moments=] and [=unsafe moments=] represent points in time, which means they can't be directly stored as numbers. Implementations will usually represent a [=moment=] as a [=duration=] from some other fixed point in time, but specifications ought to deal in the [=moments=] themselves. </p> <p> A <dfn data-export="">duration</dfn> is the distance from one [=moment=] to another from the same [=clock=]. Neither endpoint can be an [=unsafe moment=] so that both [=durations=] and differences of [=durations=] mitigate the concerns in [[[#clock-resolution]]]. [=Durations=] are measured in milliseconds, seconds, etc. Since all [=clocks=] attempt to count at the same rate, [=durations=] don't have an associated [=clock=], and a [=duration=] calculated from two [=moments=] on one clock can be added to a [=moment=] from a second [=clock=], to produce another [=moment=] on that second [=clock=]. </p> <p> The <dfn data-export="">duration from</dfn> |a:moment| to |b:moment| is the result of the following algorithm: </p> <ol class="algorithm"> <li>Assert: |a| was created by the same [=clock=] as |b|. </li> <li>Assert: Both |a| and |b| are [=coarsened moments=]. </li> <li>Return the amount of time from |a| to |b| as a [=duration=]. If |b| came before |a|, this will be a negative [=duration=]. </li> </ol> <p> [=Durations=] can be used implicitly as {{DOMHighResTimeStamp}}s. To <dfn>implicitly convert a duration to a timestamp</dfn>, given a [=duration=] |d:duration|, return the number of milliseconds in |d|. </p> </section> </section> <section id="sec-tools"> <h3> Tools for Specification Authors </h3> <p> For measuring time within a single page (within the context of a single [=environment settings object=]), use the |settingsObject:environment settings object|'s <dfn data-export="" data-dfn-for= "environment settings object">current relative timestamp</dfn>, defined as the [=duration from=] |settingsObject|'s [=environment settings object/time origin=] to the |settingsObject|'s [=environment settings object/current monotonic time=]. This value can be exposed directly to JavaScript using the [=duration=]'s [=implicitly convert a duration to a timestamp|implicit conversion=] to {{DOMHighResTimeStamp}}. </p> <p> For measuring time within a single UA execution when an [=environment settings object=]'s [=environment settings object/time origin=] isn't an appropriate base for comparison, create [=moments=] using an [=environment settings object=]'s [=environment settings object/current monotonic time=]. An [=environment settings object=] |settingsObject:environment settings object|'s <dfn data-export="" data-dfn-for="environment settings object">current monotonic time</dfn> is the result of the following steps: </p> <ol> <li>Let |unsafeMonotonicTime:unsafe moment on the monotonic clock| be the [=monotonic clock=]'s [=monotonic clock/unsafe current time=]. </li> <li>Return the result of calling [=coarsen time=] with |unsafeMonotonicTime| and |settingsObject|'s [=environment settings object/cross-origin isolated capability=]. </li> </ol> <p> [=Moments=] from the [=monotonic clock=] can't be directly represented in JavaScript or HTTP. Instead, expose a [=duration=] between two such [=moments=]. </p> <p> For measuring time across multiple UA executions, create [=moments=] using the [=current coarsened wall time=] or (if you need higher precision in [=environment settings object/cross-origin isolated capability|cross-origin-isolated contexts=]) an [=environment settings object=]'s [=environment settings object/current wall time=]. The <dfn data-export="" id="dfn-current-wall-time">current coarsened wall time</dfn> is the result of calling [=coarsen time=] with the [=wall clock=]'s [=wall clock/unsafe current time=]. </p> <p> An [=environment settings object=] |settingsObject:environment settings object|'s <dfn data-export="" data-dfn-for= "environment settings object" id="dfn-eso-current-wall-time">current wall time</dfn> is the result of the following steps: </p> <ol> <li>Let |unsafeWallTime:unsafe moment on the wall clock| be the [=wall clock=]'s [=wall clock/unsafe current time=]. </li> <li>Return the result of calling [=coarsen time=] with |unsafeWallTime| and |settingsObject|'s [=environment settings object/cross-origin isolated capability=]. </li> </ol> <p class="advisement"> When using [=moments=] from the [=wall clock=], be sure that your design accounts for situations when the user adjusts their clock either forward or backward. </p> <p> [=Moments=] from the [=wall clock=] can be represented in JavaScript by passing the number of milliseconds from the [=Unix epoch=] to that [=moment=] into the {{Date}} constructor, or by passing the number of nanoseconds from the [=Unix epoch=] to that [=moment=] into the <a data-link-type="idl" data-cite= "Temporal#sec-temporal-instant-constructor">Temporal.Instant</a> constructor. </p> <p class="advisement tracking-vector"> Avoid sending similar representations between computers, as doing so will expose the user's clock skew, which is a [=tracking vector=]. Instead, use an approach similar to [=monotonic clock=] [=moments=] of sending a duration between two [=moments=]. </p> <section id="sec-tools-examples"> <h3> Examples </h3> <div class="example" id="example-current-relative-timestamp"> <p> The time a DOM event happens can be reported using: </p> <ol start="7"> <li>Initialize |event|'s {{Event/timeStamp}} attribute to [=this=]'s [=relevant settings object=]'s [=environment settings object/current relative timestamp=]. </li> </ol> </div> <div class="example" id="example-current-monotonic-time"> <p> The age of an error report can be computed using: </p> <ol start="7"> <li>Initialize |report|'s <a data-cite= "reporting#report-timestamp">generation time</a> to |settings|' [=environment settings object/current monotonic time=]. </li> </ol> <p> Later: </p> <ol start="2"> <li>Let |data| be a map with the following key/value pairs: <dl> <dt> age </dt> <dd> The number of milliseconds between |report|'s <a data-cite= "reporting#report-timestamp">generation time</a> and |context|'s [=relevant settings object=]'s [=environment settings object/current monotonic time=], rounded to the nearest integer. </dd> <dd> ... </dd> </dl> </li> </ol> </div> <div class="example" id="example-current-wall-time"> <p> Multi-day attribution report expirations can be handled as: </p> <ol start="2"> <li>Let |source| be a new attribution source struct whose items are: <dl> <dt> ... </dt> <dt> source time </dt> <dd> |context|'s [=environment settings object/current wall time=] </dd> <dt> expiry </dt> <dd> <a data-cite="html#parse-a-duration-string">parse a duration string</a> from <code>|value|["expiry"]</code> </dd> </dl> </li> </ol> <p> Days later: </p> <ol start="2"> <li>If |context|'s [=environment settings object/current wall time=] is less than |source|'s source time + |source|'s expiry, send a report. </li> </ol> </div> </section> </section> <section id="sec-time-origin"> <h3> Time Origin </h3> <p> The <dfn data-export="">Unix epoch</dfn> is the [=moment=] on the [=wall clock=] corresponding to 1 January 1970 00:00:00 UTC. </p> <p> Each group of [=environment settings objects=] that could possibly communicate in any way has an <dfn>estimated monotonic time of the Unix epoch</dfn>, a [=moment=] on the [=monotonic clock=], whose value is initialized by the following steps: </p> <ol data-algorithm= "initialize the estimated monotonic time of the Unix epoch"> <li>Let |wall time:unsafe moment on the wall clock| be the [=wall clock=]'s [=wall clock/unsafe current time=]. </li> <li>Let |monotonic time:unsafe moment on the monotonic clock| be the [=monotonic clock=]'s [=monotonic clock/unsafe current time=]. </li> <li>Let |epoch time:unsafe moment on the monotonic clock| be <code>|monotonic time| - (|wall time| - [=Unix epoch=])</code> </li> <li>Initialize the [=estimated monotonic time of the Unix epoch=] to the result of calling [=coarsen time=] with |epoch time|. </li> </ol> <div class="issue"> The above set of settings-objects-that-could-possibly-communicate needs to be specified better. It's similar to <a data-cite= "html/browsers.html#familiar-with">familiar with</a> but includes {{Worker}}s. </div> <p> Performance measurements report a [=duration=] from a [=moment=] early in the initialization of a relevant [=environment settings object=]. That [=moment=] is stored in that settings object's [=environment settings object/time origin=]. </p> <p> To <dfn>get time origin timestamp</dfn>, given a [=/global object=] |global:global object|, run the following steps, which return a [=duration=]: </p> <ol> <li> <p> Let |timeOrigin:moment on the monotonic clock| be |global|'s [=relevant settings object=]'s [=environment settings object/time origin=]. </p> <p class="note"> In {{Window}} contexts, this value represents the time when [=navigate|navigation has started=]. In {{Worker}} and {{ServiceWorker}} contents, this value represent the time when the [=run a worker|worker is run=]. [[service-workers]] </p> </li> <li>Return the [=duration from=] the [=estimated monotonic time of the Unix epoch=] to |timeOrigin|. </li> </ol> <p class="note"> The value returned by [=get time origin timestamp=] is approximately the time after the [=Unix epoch=] that |global|'s [=environment settings object/time origin=] happened. It may differ from the value returned by <a>Date.now()</a> executed at the time origin, because the former is recorded with respect to a <a>monotonic clock</a> that is not subject to system and user clock adjustments, clock skew, and so on. </p> <div data-algorithm="coarsen time"> The <dfn data-export="">coarsen time</dfn> algorithm, given an [=unsafe moment=] |timestamp:unsafe moment| on some [=clock=] and an optional boolean |crossOriginIsolatedCapability:boolean| (default false), runs the following steps: <ol> <li>Let |time resolution:duration| be 100 microseconds, or a higher <a>implementation-defined</a> value. </li> <li>If |crossOriginIsolatedCapability| is true, set |time resolution| to be 5 microseconds, or a higher <a>implementation-defined</a> value. </li> <li>In an <a>implementation-defined</a> manner, coarsen and potentially jitter |timestamp| such that its resolution will not exceed |time resolution|. </li> <li>Return |timestamp| as a [=moment=]. </li> </ol> </div> <div data-algorithm="relative high resolution time"> The <dfn data-export="">relative high resolution time</dfn> given an [=unsafe moment=] from the [=monotonic clock=] |time:unsafe moment on the monotonic clock| and a [=Realm/global object=] |global:global object|, is the [=duration=] returned from the following steps: <ol> <li>Let |coarse time:moment on the monotonic clock| be the result of calling [=coarsen time=] with |time| and |global|'s [=relevant settings object=]'s [=environment settings object/cross-origin isolated capability=]. </li> <li>Return the [=relative high resolution coarse time=] for |coarse time| and |global|. </li> </ol>The <dfn data-export="">relative high resolution coarse time</dfn> given a [=moment=] from the [=monotonic clock=] |coarseTime:moment on the monotonic clock| and a [=Realm/global object=] |global:global object|, is the [=duration from=] |global|'s [=relevant settings object=]'s [=environment settings object/time origin=] to |coarseTime|. </div> <p> The <dfn data-export="">current high resolution time</dfn> given a [=/global object=] |current global:global object| must return the result of [=relative high resolution time=] given [=unsafe shared current time=] and |current global|. </p> <p> The <dfn data-export="">coarsened shared current time</dfn> given an optional boolean |crossOriginIsolatedCapability:boolean| (default false), must return the result of calling [=coarsen time=] with the [=unsafe shared current time=] and |crossOriginIsolatedCapability|. </p> <p> The <dfn data-export="">unsafe shared current time</dfn> must return the [=monotonic clock/unsafe current time=] of the <a>monotonic clock</a>. </p> </section> <section id="sec-domhighrestimestamp"> <h3> The <dfn data-export="">DOMHighResTimeStamp</dfn> typedef </h3> <p> The {{DOMHighResTimeStamp}} type is used to store a [=duration=] in milliseconds. Depending on its context, it may represent the [=moment=] that is this [=duration=] after a base [=moment=] like a [=environment settings object/time origin=] or the [=Unix epoch=]. </p> <pre class="idl"> typedef double DOMHighResTimeStamp; </pre> <p> A {{DOMHighResTimeStamp}} SHOULD represent a time in milliseconds accurate enough to allow measurement while preventing timing attacks - see <a href="#clock-resolution"></a> for additional considerations. </p> <p class="note"> A {{DOMHighResTimeStamp}} is a {{double}}, so it can only represent an epoch-relative time—the number of milliseconds from the [=Unix epoch=] to a [=moment=]—to a finite resolution. For [=moments=] in 2023, that resolution is approximately 0.2 microseconds. </p> </section> <section> <h3> The <dfn>EpochTimeStamp</dfn> typedef </h3> <pre class="idl"> typedef unsigned long long EpochTimeStamp; </pre> <aside class="note" title="Legacy platform feature"> <p> The use of `EpochTimeStamp`, known previously as `DOMTimeStamp`, is discouraged. Wherever possible use {{DOMHighResTimeStamp}} instead. </p> </aside> <p> A {{EpochTimeStamp}} represents an integral number of milliseconds from the [=Unix epoch=] to a given [=moment=] on the [=wall clock=], excluding leap seconds. Specifications that use this type define how the number of milliseconds are interpreted. </p> </section> <section id="sec-performance" data-dfn-for="Performance"> <h3> The <dfn>Performance</dfn> interface </h3> <pre class="idl"> [Exposed=(Window,Worker)] interface Performance : EventTarget { DOMHighResTimeStamp now(); readonly attribute DOMHighResTimeStamp timeOrigin; [Default] object toJSON(); }; </pre> <section> <h3> `now()` method </h3> <p data-tests="basic.any.html, basic.any.worker.html"> The <dfn>now()</dfn> method MUST return the number of milliseconds in the <a>current high resolution time</a> given [=this=]'s [=relevant global object=] (a [=duration=]). </p> <p data-tests= "monotonic-clock.any.html, monotonic-clock.any.worker.html"> The time values returned when calling the {{Performance/now()}} method on {{Performance}} objects with the same [=environment settings object/time origin=] MUST use the same [=monotonic clock=]. The difference between any two chronologically recorded time values returned from the {{Performance/now()}} method MUST never be negative if the two time values have the same [=environment settings object/time origin=]. </p> </section> <section> <h3> `timeOrigin` attribute </h3> <p data-tests="timeOrigin.html, window-worker-timeOrigin.window.html"> The <dfn>timeOrigin</dfn> attribute MUST return the number of milliseconds in the [=duration=] returned by [=get time origin timestamp=] for the <a>relevant global object</a> of [=this=]. </p> <p data-tests="test_cross_frame_start.html"> The time values returned when getting {{Performance}}.{{Performance/timeOrigin}} MUST use the same [=monotonic clock=] that is shared by [=environment settings object/time origins=], and whose reference point is the [[ECMA-262]] <a data-cite="ECMA-262#sec-time-values-and-time-range">time</a> definition - see [[[#sec-security]]]. </p> </section> <section> <h3> `toJSON()` method </h3> <p data-tests="performance-tojson.html"> When <dfn>toJSON()</dfn> is called, run [[WEBIDL]]'s <a>default toJSON steps</a>. </p> </section> </section> <section> <h2> Extensions to `WindowOrWorkerGlobalScope` mixin </h2> <section data-dfn-for="WindowOrWorkerGlobalScope"> <h3> The <code>performance</code> attribute </h3> <p> The <dfn>performance</dfn> attribute on the interface mixin {{WindowOrWorkerGlobalScope}} allows access to performance related attributes and methods from the [=Realm/global object=]. </p> <pre class="idl"> partial interface mixin WindowOrWorkerGlobalScope { [Replaceable] readonly attribute Performance performance; }; </pre> </section> </section> <section id="sec-security"> <h3> Security Considerations </h3> <section> <h3> Clock resolution </h3> <p> Access to accurate timing information, both for measurement and scheduling purposes, is a common requirement for many applications. For example, coordinating animations, sound, and other activity on the page requires access to high-resolution time to provide a good user experience. Similarly, measurement enables developers to track the performance of critical code components, detect regressions, and so on. </p> <p> However, access to the same accurate timing information can sometimes be also used for malicious purposes by an attacker to guess and infer data that they can't see or access otherwise. For example, cache attacks, statistical fingerprinting and micro-architectural attacks are a privacy and security concern where a malicious web site may use high resolution timing data of various browser or application-initiated operations to differentiate between subset of users, identify a particular user or reveal unrelated but same-process user data - see [[?CACHE-ATTACKS]] and [[SPECTRE]] for more background. </p> <p data-tests="timing-attack.html"> This specification defines an API that provides sub-millisecond time resolution, which is more accurate than the previously available millisecond resolution exposed by {{EpochTimeStamp}}. However, even without this new API an attacker may be able to obtain high-resolution estimates through repeat execution and statistical analysis. </p> <p> To ensure that the new API does not significantly improve the accuracy or speed of such attacks, the minimum resolution of the {{DOMHighResTimeStamp}} type should be inaccurate enough to prevent attacks. </p> <p> Where necessary, the user agent should set higher resolution values to |time resolution| in [=coarsen time=]'s processing model, to address privacy and security concerns due to architecture or software constraints, or other considerations. </p> <p> In order to mitigate such attacks user agents may deploy any technique they deem necessary. Deployment of those techniques may vary based on the browser's architecture, the user's device, the content and its ability to maliciously read cross-origin data, or other practical considerations. </p> <p> These techniques may include: </p> <ul> <li>Resolution reduction. </li> <li>Added jitter. </li> <li>Abuse detection and/or API call throttling. </li> </ul> <p> Mitigating such timing side-channel attacks entirely is practically impossible: either all operations would have to execute in a time that does not vary based on the value of any confidential information, or the application would need to be isolated from any time-related primitives (clock, timers, counters, etc). Neither is practical due to the associated complexity for the browser and application developers and the associated negative effects on performance and responsiveness of applications. </p> <div class="note"> Clock resolution is an unsolved and evolving area of research, with no existing industry consensus or definitive set of recommendations that applies to all browsers. To track the discussion, refer to <a href="https://github.com/w3c/hr-time/issues/79">Issue 79</a>. </div> </section> <section> <h3> Clock drift </h3> <p> This specification also defines an API that provides sub-millisecond time resolution of the zero time of the time origin, which requires and exposes a <a>monotonic clock</a> to the application, and that must be shared across all the browser contexts. The <a>monotonic clock</a> does not need to be tied to physical time, but is recommended to be set with respect to the [[ECMA-262]] definition of <a>time</a> to avoid exposing new fingerprint entropy about the user — e.g. this time can already be easily obtained by the application, whereas exposing a new logical clock provides new information. </p> <p> However, even with the above mechanism in place, the <a>monotonic clock</a> may provide additional <a href= "https://en.wikipedia.org/wiki/Clock_drift">clock drift</a> resolution. Today, the application can timestamp the time-of-day and monotonic time values (via <a>Date.now()</a> and {{Performance/now()}}) at multiple points within the same context and observe drift between them—e.g. due to automatic or user clock adjustments. With the {{Performance/timeOrigin}} attribute, the attacker can also compare the [=environment settings object/time origin=], as reported by the <a>monotonic clock</a>, against the current time-of-day estimate of the [=environment settings object/time origin=] (i.e. the difference between `performance.timeOrigin` and `Date.now() - performance.now()`) and potentially observe clock drift between these clocks over a longer time period. </p> <p> In practice, the same time drift can be observed by an application across multiple navigations: the application can record the logical time in each context and use a client or server time synchronization mechanism to infer changes in the user's clock. Similarly, lower-layer mechanisms such as TCP timestamps may reveal the same high-resolution information to the server without the need for multiple visits. As such, the information provided by this API should not expose any significant or previously unavailable entropy about the user. </p> </section> </section> <section id="sec-privacy"> <h3> Privacy Considerations </h3> <p> The current definition of [=environment settings object/time origin=] for a {{Document}} exposes the total time of cross-origin redirects prior to the request arriving at the document's origin. This exposes cross-origin information, however it's not yet decided how to mitigate this without causing major breakages to performance metrics. </p> <p> To track the discussion, refer to <a href= "https://github.com/w3c/navigation-timing/issues/160">Navigation Timing Issue 160</a>. </p> </section> <section id="conformance"> <p> Some conformance requirements are phrased as requirements on attributes, methods or objects. Such requirements are to be interpreted as requirements on user agents. </p> </section> <section id="index"></section> <section id="idl-index"></section> <section class="appendix"> <h2> Acknowledgments </h2> <p> Thanks to Arvind Jain, Angelos D. Keromytis, Boris Zbarsky, Jason Weber, Karen Anderson, Nat Duca, Philippe Le Hegaret, Ryosuke Niwa, Simha Sethumadhavan, Todd Reifsteck, Tony Gentilcore, Vasileios P. Kemerlis, Yoav Weiss, and Yossef Oren for their contributions to this work. </p> </section> </body> </html>

Pages: 1 2 3 4 5 6 7 8 9 10