Comparison of sequential vs parallel Azure Read API processing time

compare.py

import requests
import io
import time

n = 50

# Enter your resource details here
url = "https://xxxxxxx.cognitiveservices.azure.com/vision/v3.2/read/analyze?language=en&pages=1&readingOrder=natural"
key = "xxxxxxx"

headers = {
    "Content-Type": "application/json",
    "Ocp-Apim-Subscription-Key": key
}

def start_ocr(url, headers):
    json = {
        "url": "https://docs.microsoft.com/en-us/azure/cognitive-services/computer-vision/images/description.png"
    }
    start_time = time.time()
    result = requests.post(url, json=json, headers=headers)
    assert(result.status_code == 202)
    duration = time.time() - start_time
    if (duration < 0.1):
        time.sleep(0.1-duration) # Ensure we stay under 10 TPS
    return result.headers['Operation-Location']
    
def wait_for_results(url, headers):
    status = "running"
    while(status != "succeeded"):
        start_time = time.time()
        status = requests.get(url, headers=headers).json()['status']
        duration = time.time() - start_time
        if (duration < 0.1):
            time.sleep(0.1-duration) # Ensure we stay under 10 TPS

### Run sequential test ####
start_time_seq = time.time()
for i in range(0,n):
    r_url = start_ocr(url, headers)
    wait_for_results(r_url, headers)
    # Do something with the results :)
duration_seq = time.time() - start_time_seq
print(f"Sequential batch of {n} took {duration_seq} seconds")

#### Run parallel test ####
start_time_parallel = time.time()
urls = []
for i in range(0,n):
    urls.append(start_ocr(url, headers))

assert(len(urls) == n)
for url in urls:
    wait_for_results(url, headers)
    # Do something with the results :)
duration_parallel = time.time() - start_time_parallel
print(f"Parallel batch of {n} took {duration_parallel} seconds ({int(100*duration_parallel/duration_seq)}%)")

Output for westeurope:

Sequential batch of 50 took 36.734880208969116 seconds
Parallel batch of 50 took 11.86894941329956 seconds (32%)