gannon@Indiana.edu
Dennis.gannon@outlook.com
Microservices and Streaming Data:
Scaling ML in the Cloud
compute
compute
compute
compute
compute
compute
compute
compute
compute
compute
compute
compute
Marathon
Master
node
Master
Backup
Worker
node
Worker
node
Worker
node
Mesos
Zookeeper
*
*
classify Push to table
classify Push to table
classify Push to table
classify Push to table
classify Push to table
classify Push to table
classify Push to table
te te te
ta ta
classify Push to table classify Push to table
𝑡 𝑒 = time spent pulling events and pushing
notifications to RabbitMQ
𝑡 𝑐 = time classify documents
𝑡 𝑤 = time to invoke the web service
𝑡 𝑎 = time to put final results into the
Azure table service
𝑡 𝑠𝑒𝑞 = 𝑓 𝑡 𝑒 , 𝑡 𝑎
𝑡 𝑝𝑎𝑟 = 𝑡 𝑐 + 𝑡 𝑤
𝑇𝑠𝑒𝑟𝑖𝑎𝑙 = 𝑡 𝑠𝑒𝑞 + 𝑡 𝑝𝑎𝑟
𝑇𝑝𝑎𝑟 𝑛 = 𝑡 𝑠𝑒𝑞 +
𝑡 𝑝𝑎𝑟
𝑛
𝑆 𝑛 =
𝑇𝑠𝑒𝑟𝑖𝑎𝑙
𝑇𝑝𝑎𝑟 𝑛
=
1 +
𝑡 𝑝𝑎𝑟
𝑡 𝑠𝑒𝑞
1 +
𝑡 𝑝𝑎𝑟
𝑛 ∗ 𝑡 𝑠𝑒𝑞
lim
𝑛→∞
𝑆(𝑛) = 1 +
𝑡 𝑝𝑎𝑟
𝑡 𝑠𝑒𝑞
1.00 2.02 3.53 7.91 12.33 13.05
Microservice performance-b
Microservice performance-b
Microservice performance-b

Microservice performance-b