Skip to content

Latest commit

 

History

History
68 lines (52 loc) · 3.56 KB

Sizing.md

File metadata and controls

68 lines (52 loc) · 3.56 KB
Raw

Infrastructure sizing

Throughput calculation

Average throughput per second per patient:

Sensor Signals per second
Heart rate 2
Blood pressure 0,000277777777777778
Oxygen level 0,2
Blood sugar 0,00833333333333333
Respiration 1
ECG 1
Body temperature 0,00333333333333333
Sleep status 0,00833333333333333
Total 4,22027777777778

If we assume, that maximum patients number is 500 and the single signal can utilize 1kB, than:

Throughput type Value Total (req/s) Throughput (MB/s)
AVERAGE 4,2 * 500 2100 2,1 MB/s
PEAK 8 * 500 4000 4 MB/s

This values shows us, that system MUST be ready to process 4000 signals in one second and network must be prepared to transfer at least 4MB/s.

Estimated database write time utilization.

In order to validate a storing process, we have to examine a specific database. For this Fitness Function, it has been chosen the Influx DB, but the final decision which database to use, will be taken later, Based on following benchmark we can assume that:

  • Influx DB can save at least 237871 events/s
  • In our case, we need to store 4000 events/s, so the storing process shouldn't be longer than 16ms

Publishing to Message Streaming Platform

As the previous calculation, here was chosen Apache Kafka solution as a streaming platform and again - the final decision will be taken later. Based on AWS benchmark

considering the machine 16GB RAM, 4CPU and fast SSD disk (equivalent of m5.xlarge) the Kafka needs 4ms to publish 4MB of data in second.

LAN throughput

If we assume that the hospital has a 1Gb/s LAN, transmitting 4MB of data should take approximately 32 ms.

Vital Sign Streamer

If we assume that we have 3 instances of Vital Sign Streamer services:

  • Each instance must handle approximately 8.33 connections (25 nurse stations / 3 services)
  • Let's assume that each connection handles requests on separate threads (without considering processor context switching).
    • 1 connection should handle 20 patients, which means 160 signals per second (20 connections x 8 sensors).
    • If we assume that processing 1 signal takes 2 milliseconds, then the time needed to process signals on 1 connection is 320ms.

Vital Sign Recorder

To simplify calculations, we can assume that the amount of data is exactly the same as the Streamer process. Then, we can simply state that the same set of 3 Vital Sign Recorder services also requires 320ms.

Final calculation (from sensor to nurse station)

Step Total Time
Vital Sign Recorder 320ms
Vital Sign Streamer 320ms
Save to DB 16ms
Send to Streaming Platform 5ms
Local Network LAN Data Transfer 32ms
Total 693ms