Announcement: All noncommercial projects registered to use Earth Engine before April 15, 2025 must verify noncommercial eligibility to maintain Earth Engine access.
Stay organized with collections
Save and categorize content based on your preferences.
Earth Engine is built on top of Google's tools and services for performing
computations at a massive scale. To make it easy to run large geospatial
analyses, the Earth Engine platform and API hide much of the complexity of the
underlying parallel-processing infrastructure.
EECUs
Overview
An Earth Engine Compute Unit (EECU) is a mechanism for representing an amount of
instantaneous processing power. Earth Engine tracks the total computational
footprint of tasks as a function of their EECU usage through time (EECU-seconds,
EECU-hours, etc.). Because Google has many different types of processor cores,
architectures, etc., EECUs are a useful abstraction for talking about
computational power.
Motivation
EE users often want to make estimates about the amount of processing power
required for their workflows, and EECUs provide a consistent metric for making
comparisons.
Comparison with CPU metrics
The number, type and architecture of machines working on a particular result can
change over time. Because different physical cores can have different
performance characteristics, Earth Engine abstracts all processing using EECUs.
An EECU-hour (or any other unit of EECU-time) doesn't correspond to a wall clock
time, so a job which consumes 10 EECU-hours may have an observed runtime of just
a few minutes.
Stability and predictability
Sending the same (or similar) requests to Earth Engine can sometimes result in
very different amounts of computation. Common drivers of differences include:
caching, such as reusing the results of previous computations (including
partial or intermediate results)
different underlying data, such as varying numbers of satellite images,
geometries of different complexity, etc.
algorithm changes on the EE platform, including performance
optimizations, bugfixes, etc.
changes to client libraries, particularly if you depend on other users'
EE code or packages
Earth Engine doesn't provide performance metrics for failed requests/tasks,
since these numbers would be inaccurate or misleading. As an example, if a job
fails because a worker task became unresponsive, that worker's processing
consumption wouldn't be able to factor into the total.
Profiler
The profiler provides information about EECU-time and memory usage (per
algorithm and asset) resulting from the computation performed while it's
enabled. Each row in the profiler output corresponds to an algorithm,
computation, asset load or overhead operation as described in the 'Description'
column. The columns in the profiler are:
Description
A textual description of the computation, algorithm, asset load or
overhead operation being profiled.
Count
An indicator proportional to the number of times the operation described
in 'Description' was invoked.
Compute
An indicator of EECU-time taken by the operation(s).
Current Mem
This column appears only if there was an error because the script
used too much memory. It shows the amount of memory in use on any single
compute node at the moment the error occurred.
Peak Mem
Maximum memory used on any single compute node for the operation.
Enabling the profiler
Code Editor
Use the "Run with Profiler" button, as described in the Code Editor
guide.
Python
Include the following code in your Python script to enable the profiler:
Here is a suggestion for turning the profile string into a table for easier
analysis in Colab and Jupyter Notebooks (note that this is just one approach
and may not be suitable for all cases):
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-01-21 UTC."],[[["\u003cp\u003eEarth Engine Compute Units (EECUs) represent the amount of processing power used in Earth Engine, allowing users to estimate and compare computational needs for geospatial analysis.\u003c/p\u003e\n"],["\u003cp\u003eEECU-time is an abstract measure of work and doesn't directly equate to CPU or wall clock time due to Earth Engine's dynamic resource allocation.\u003c/p\u003e\n"],["\u003cp\u003eWhile Earth Engine strives for computational stability, factors like caching, data variations, algorithm updates, and client library changes can influence processing requirements.\u003c/p\u003e\n"],["\u003cp\u003eUsers can leverage the Earth Engine profiler to gain insights into EECU and memory usage for specific algorithms, asset loads, and operations within their scripts.\u003c/p\u003e\n"]]],[],null,["Earth Engine is built on top of Google's tools and services for performing\ncomputations at a massive scale. To make it easy to run large geospatial\nanalyses, the Earth Engine platform and API hide much of the complexity of the\nunderlying parallel-processing infrastructure.\n\nEECUs\n\nOverview **Key Term:** *EECU* - Earth Engine Compute Unit, an abstraction of computing power.\n\nAn Earth Engine Compute Unit (EECU) is a mechanism for representing an amount of\ninstantaneous processing power. Earth Engine tracks the total computational\nfootprint of tasks as a function of their EECU usage through time (EECU-seconds,\nEECU-hours, etc.). Because Google has many different types of processor cores,\narchitectures, etc., EECUs are a useful abstraction for talking about\ncomputational power.\n\nMotivation\n\nEE users often want to make estimates about the amount of processing power\nrequired for their workflows, and EECUs provide a consistent metric for making\ncomparisons.\n\nComparison with CPU metrics **Note:** While EECU-seconds measure the amount of work that Earth Engine is performing, EECU-seconds don't correspond directly to CPU-seconds or wall clock seconds due to the fully managed nature of the service.\n\nThe number, type and architecture of machines working on a particular result can\nchange over time. Because different physical cores can have different\nperformance characteristics, Earth Engine abstracts all processing using EECUs.\nAn EECU-hour (or any other unit of EECU-time) doesn't correspond to a wall clock\ntime, so a job which consumes 10 EECU-hours may have an observed runtime of just\na few minutes.\n\nStability and predictability\n\nSending the same (or similar) requests to Earth Engine can sometimes result in\nvery different amounts of computation. Common drivers of differences include:\n\n- **caching**, such as reusing the results of previous computations (including partial or intermediate results)\n- **different underlying data**, such as varying numbers of satellite images, geometries of different complexity, etc.\n- **algorithm changes** on the EE platform, including performance optimizations, bugfixes, etc.\n- **changes to client libraries**, particularly if you depend on other users' EE code or packages\n\nBenchmarks\n\nExplore [sample Earth Engine computation benchmarks](./computation_benchmarks).\n\nMetrics for failed requests\n\nEarth Engine doesn't provide performance metrics for failed requests/tasks,\nsince these numbers would be inaccurate or misleading. As an example, if a job\nfails because a worker task became unresponsive, that worker's processing\nconsumption wouldn't be able to factor into the total.\n\nProfiler\n\nThe profiler provides information about EECU-time and memory usage (per\nalgorithm and asset) resulting from the computation performed while it's\nenabled. Each row in the profiler output corresponds to an algorithm,\ncomputation, asset load or overhead operation as described in the 'Description'\ncolumn. The columns in the profiler are:\n\nDescription\n: A textual description of the computation, algorithm, asset load or\n overhead operation being profiled.\n\nCount\n: An indicator proportional to the number of times the operation described\n in 'Description' was invoked.\n\nCompute\n: An indicator of EECU-time taken by the operation(s).\n\nCurrent Mem\n\n: This column appears only if there was an error because the script\n\n used too much memory. It shows the amount of memory in use on any single\n compute node at the moment the error occurred.\n\nPeak Mem\n\n: Maximum memory used on any single compute node for the operation.\n\nEnabling the profiler \n\nCode Editor\n\nUse the \"Run with Profiler\" button, as described in the [Code Editor\nguide](/earth-engine/guides/playground#profiler).\n\nPython\n\nInclude the following code in your Python script to enable the profiler: \n\n with ee.profilePrinting():\n print(ee.Number(3.14).add(0.00159).getInfo())\n\nThe profile will be printed when the context ends, whether or not any error\noccurred within the context.\n\nTo capture the profile as a string, write the profile to a string buffer: \n\n import io\n\n out = io.StringIO()\n with ee.profilePrinting(destination=out) as p:\n print(ee.Number(3.14).add(0.00159).getInfo())\n\n print('Output:')\n print(out.getvalue())\n\nHere is a suggestion for turning the profile string into a table for easier\nanalysis in Colab and Jupyter Notebooks (note that this is just one approach\nand may not be suitable for all cases): \n\n import re\n import pandas as pd\n\n lines = out.getvalue().split('\\n')\n\n column_names = re.split(r'\\s{1,}', lines[0])\n column_names = [name.strip() for name in column_names if name.strip()]\n\n data = [\n [element for element in re.split(r'\\s{2,}', line) if element.strip()]\n for line in lines[1:-1]\n ]\n\n df = pd.DataFrame(data, columns=column_names)\n display(df)"]]