benchlog 0.3

Log CPU, GPU and Memory usage for a project overtime and send to server

Benchlog Package

Benchlog allows a user to record the runtime and resource allocation of their program overtime.

Changelog

Version 0.3

Features

• Added logging of accuracy
• Added unique filename generation (using date and time)
• [Documentation] Added a changelog
• Allow local saving and sending to server with setHost(HOST, local)
• Added GPU logging of usage, memory, memory utilization, gpu UUID and name with enableGPU()
• Prevent users that do not have 'nvidia-smi' installed from enabling GPU

Bug Fixes

• Fixed bug where sendData would only save to telemetry.json on error instead of calling saveData
• Prevented user from calling end() without calling start()
• Prevented user from calling end() twice
• Call for variable deviceID did not exist in enableGPU(), use self.gpuObj instead
• Fixed bug where we called saveData() instead of self.saveData()

Installation

pip install benchlog

Use

Benchlog Arguments

Argument	Type	Description
projectname	string	The name of your project
iterationNum	integer	Number of iteration your code uses (Used to record progress)
Opt: featureList	[string]	Record list of features used for project
Opt: quiet	integer	Set to 1 to disable logging, default: 0

Setup

1. Import BenchLog from benchlog import BenchLog

2. Declare new instance of BenchLog E.g. logging = BenchLog('Test Project', 1024, ['tensorflow', 'featureB'], quiet1]

3. Optional: Declare a host to send the telemetry data to. If no host is stored or data fails to send we will store the data in a file called telemetry.json in the directory of the project. If you would like to enable local and external logging, set the second argument to 1. setHost(ipaddress, [OPTIONAL:1 to enable local logging, 0 default]) E.g logging.setHost('http://localhost:3000',1)

4. Optional: GPU Logging - Enable GPU Logging by calling logging.enableGPU() If you have more than one GPU refer to the enableGPU section

5. Start logging by calling logging.start()

6. Call logging.log(iterationNum) every once in awhile to log progress.

7. End logging by calling logging.end()

Example Code

from benchlog import BenchLog
logging = BenchLog('My Project', 10000, ['FeatureA','FeatureB'])
logging.setHost('http://localhost:3000')
array =
logging.enableGPU(0)
logging.start()
for i in range(1,10000):
    if(i % 1000 == 0):
        #Do Something here
        logging.log(i)
logging.end()

enableGPU()

logging.enableGPU() uses GPUtil in order to gather gpu information. This will only work with Nvidia GPUs.

If you have multiple GPUs you can pass in the index of the GPU you want to use. You can find which GPU you want to log by calling nvidia-smi and returning the GPU you want to use. E.g logging.enableGPU(1)

GPU logging records three things.

GpuMemUsage - This is the amount of memory being used by the GPU GpuMemUtil - This is the memory activity (i/o) within GPU memory GpuUsage - This is the standard gpu load

Example telemetry output

{"cores":"4","runTime":"1.506113","size":"10000","features":["testa","testb"],"projectName":"test","infoRunTime":["0.177929","0.345282","0.511437","0.685919","0.850231","1.011178","1.182584","1.345929","1.505748","1.506086"],"infoCpuUsage":["13.2","32.4","30.3","35.7","30.8","28.1","36.2","26.2","28.1","0.0"],"infoMemoryUsage":["452567040","852570112","1252573184","1652576256","2052579328","2452582400","2852585472","3252588544","3652591616","3652591616"],"infoProgress":["0.1","0.2","0.3","0.4","0.5","0.6","0.7","0.8","0.9","1.0"]}

Future To Do

• Do not overwrite old telemetry data
• GPU Utilization
• Local Visualization
• Web UI to view results
• Server
• Change progress to 0-100