Skip to content

Latest commit

 

History

History
73 lines (50 loc) · 3.46 KB

research-calculations.md

File metadata and controls

73 lines (50 loc) · 3.46 KB
Raw

Calculations

The calculations below are based on the concurrent connections limit showing that a messaging server is limited to 64 k connections. The calculation tells us that 160 messaging servers are needed in order to support 10 mln active users. We would consider 256 bytes message size.

Daily 24 hour usage

Calculating the daily usage with 64 mln users spread throughout the day each of which sends 128 messages per day gives us 95 000 messages/s for the cell and 23 MB/s for the cell with 0.15 MB/s per messaging server.

Peak 1 hour usage

Calculating a peak usage for 1 hour daily where 80% of the maximum users - 8 mln active users send 50% of their daily messages - 64 messages we get 142 200 messages/s for the cell and 35 MB/s for the cell with 0.22 MB/s per messaging server.

Conclusion

These numbers do not take into account the security and transport data overhead. Additionally, this traffic would be multiplied. For example sending a message would require that data to be passed between different layers, possibly to multiple recipients. Numbers are not small when we look at the system as a whole. But for a single messaging server the throughput is tiny. The system is limited by how many concurrent connections a messaging server can handle. That means we would need a lot of messaging servers, linearly scalable technologies and we need to support a high level of concurrency. The usage of Kafka, Cassandra and the .NET async programming model make a good start.

System limits

  • Targets
    • 10 000 000 max users at the same time
    • At least 4 years of message storage
    • 256 bytes per message
  • Infrastructure
    • 64 000 connections per messaging server
    • 160 messaging servers
    • 10 240 000 max active users are supported
  • Time
    • 1 hour = 60 minutes * 60 seconds = 3 600 seconds
    • 1 day = 24 hours * 60 minutes * 60 seconds = 86 400 seconds

Daily

  • Users come and go throughout the day
  • 64 000 000 active users daily
  • 128 messages per user daily

Messages

  • 64 000 000 users x 128 messages = 8 192 000 000 messages daily
  • 8 192 000 000 messages daily / 86 400 seconds = ~95 000 messages/s daily
  • 95 000 messages/s / 160 messaging servers = ~600 messages/s per messaging server daily

Throughput

  • 8 192 000 000 messages * 256 bytes = 2 097 152 000 000 bytes = 2 048 000 000 KB = 2 000 000 MB daily = ~1 953 GB daily = ~1.91 TB daily
  • 2 000 000 MB daily / 86 400 seconds = ~23 MB/s daily
  • 23 MB/s / 160 messaging servers = ~0.15 MB/s per messaging server daily

Peak

  • 1 hour peak
  • 80% of system is busy = 8 000 000 active users
  • 50% of user daily messages are sent = 64 messages for each user peak

Messages

  • 8 000 000 users * 64 messages = 512 000 000 messages peak
  • 512 000 000 peak / 3 600 seconds = ~142 200 messages/s peak
  • 142 000 messages/s / 160 messaging servers = ~900 messages/s per messaging server peak

Throughput

  • 512 000 000 messages * 256 bytes = 131 072 000 000 bytes = 128 000 000 KB = 125 000 MB peak = ~122 GB peak = ~0.12 TB peak
  • 125 000 MB peak / 3 600 seconds = ~35 MB/s peak
  • 35 MB/s / 160 messaging servers = ~0.22 MB/s per messaging server peak

Storage

  • 2 000 000 MB daily (from daily throughput)
  • 730 000 000 MB yearly
  • 2 920 000 000 MB for 4 years = 2 851 562 GB for 4 years = ~2 785 TB for 4 years = ~2.72 PT for 4 years
  • ~44 PT for 64 years
  • These calculations do not take into account data compression