TAP Precision Time APIs: Difference between revisions

From OpenCompute
Jump to navigation Jump to search
No edit summary
 
(One intermediate revision by one other user not shown)
Line 1: Line 1:
[[File:Screenshot_2020-07-01_16.35.12.png|right]]
[[File:Screenshot_2020-07-01_16.35.12.png|right]]
==Workstream #3==
==Workstream #3==
[https://www.opencompute.org/wiki/Time_Appliance_Project Time Appliances Project]  
[https://www.opencompute.org/wiki/Time_Appliance_Project Back to Time Appliances Project's wiki]


Unix clock_gettime API was designed a couple of decades ago. The time is encoded as number of units between now and a date half a century ago. In theory this is sufficient in the field of physics, and it should be sufficient any use of time as a phenomenon. For the majority of the systems, it was sufficient in practice as well. Representing the time is an offset from the epoch worked well for most of us, and for half of a century, the only change was a few improvements related the size of the unit.  
The ''clock_gettime'' API was designed a couple of decades ago. The time is encoded as number of units between now and a half of a century ago. In theory, this encoding should be sufficient for any use of the phenomenon. For the majority of the systems, representing the time as an offset from the epoch works well. The changes in the last few decades were improvements related to the size of the unit; the unit shrank from seconds to nanoseconds.  
   
   
Representing the time as just a number works well for many software systems. However, in the last two decades, to increase the efficiency of the system, we started building system where we replace communication with simple local calculation. For such systems, we need more information about the clock synchronization process. It became important when was the clock synchronized, what type of physical local clock we have, what was the last measured offset between local clock and the referent clock.
Representing the time as just a number works well for many software systems. However, in the last two decades, to increase the efficiency of the system, we started building system where we replace communication with simple local calculation. For such systems, we need more information about the clock synchronization process. It became important when was the last time when the clock was synchronized, what type of physical local clock we have, what was the last measured offset between the local clock and the referent clock.


In the Time API stream, we will start with discussing software system and services that would benefit from better clock and time API. Initially I would like to focus our discussions on defining the problem that we want to solve, any requirements and existing solutions. At some point, we could discuss building a reference implementation, but that is not the primary goal. The goal is to publish an official proposal for time and clock API that allow designing system where time is reliable fundamental building block.
In the Time API stream, we will start with discussing software system and services that benefit from better time APIs. Initially I would like to focus our discussions on defining the problem that we want to solve, what are the requirements and what solution exist today. We will focus on defining the API and perhaps, at some point, we could discuss building a reference implementation. The goal is to publish an official proposal for time and clock API that allow designing system where time is reliable fundamental building block.


==Objective==
==Objective==

Latest revision as of 23:42, 23 May 2023

Screenshot 2020-07-01 16.35.12.png

Workstream #3[edit]

Back to Time Appliances Project's wiki

The clock_gettime API was designed a couple of decades ago. The time is encoded as number of units between now and a half of a century ago. In theory, this encoding should be sufficient for any use of the phenomenon. For the majority of the systems, representing the time as an offset from the epoch works well. The changes in the last few decades were improvements related to the size of the unit; the unit shrank from seconds to nanoseconds.

Representing the time as just a number works well for many software systems. However, in the last two decades, to increase the efficiency of the system, we started building system where we replace communication with simple local calculation. For such systems, we need more information about the clock synchronization process. It became important when was the last time when the clock was synchronized, what type of physical local clock we have, what was the last measured offset between the local clock and the referent clock.

In the Time API stream, we will start with discussing software system and services that benefit from better time APIs. Initially I would like to focus our discussions on defining the problem that we want to solve, what are the requirements and what solution exist today. We will focus on defining the API and perhaps, at some point, we could discuss building a reference implementation. The goal is to publish an official proposal for time and clock API that allow designing system where time is reliable fundamental building block.

Objective[edit]

Time and clock API that allow designing system where time is reliable fundamental building block

- APIs to bring accurate time to the user space
- APIs to disseminate the time error (error bound)
- APIs to provide viability over the time synchronization process


Project Team[edit]

- Lead: Georgi Chalakov (NVIDIA)
-
-

Get Involved[edit]