How to optimize JVM to process 175k transactions per sec?

Sanhong Li, JVM at Alibaba talks about how his company managed to process giant transactions amount at the world's largest online shopping event
04 August 2017   1733

Alibaba Group is a Chinese public company working in the field of Internet commerce, the owner of B2B web portal This is one the world-largest internet companies. 

Company manages:

  • Alibaba Pictures 
  • Alipay 
  • Juhuasuan
  • eTao
  • Alibaba Cloud Computing
  • China Yahoo!

On 11.11.16, Alibaba beat it own record - company was able to process 175 000 transactions per second. A lot of Alibaba’s e-Commerce app use Java. To cater for the specific needs to run these applications, team identified the requirements and optimized these features on the customized version of HotSpot (OpenJDK based). 

At this report, Sanhong talked about: 

  • how they characterize workloads to identify specific needs;
  • optimization and customizing HotSpot for Java apps.

Alibaba's team identified three specific features useful for their needs:

  • put multiple containers into one JVM instance, which allows to deploy many small Java applications in large scale, across data centers,
  • use coroutines (from Da Vinci Machine project) to reduce context switches,
  • implement quick Java warmup to obviate the need for "warming-up" occurred in initialization phase of the eCommerce applications

Google to Release Cloud Inference API

Cloud Inference API can be used for real time big data analysis 
20 September 2018   91

Google introduced an alpha version of the service for time series analysis. It processes information about events at the time - clicks, requests, activations of IoT devices, and so on. The Cloud Inference API analyzes these data in real time, finds correlations and makes predictions based on it.

Cloud Inference API
Cloud Inference API 

Service features:

  • A simple tree-like query language that allows you to specify your own time markers.
  • Online processing of incoming data with minimal delay. Therefore, Google recommends using the API in interactive user applications.
  • Ability to process data arrays of different volumes (up to trillions of records) and work under high load (up to hundreds of thousands of requests per second).
  • Full integration with Google Cloud Storage, which provides access to the same data in different services of the platform.
  • More information about the work of the tool can be found in the documentation.

Google noted that the service will be useful for a wide range of industries. Retailers can analyze the impact of pedestrian traffic on the level of sales conversion, content providers - the popularity of materials to provide better personal recommendations.

Now the Cloud Inference API is already being used by Snap to analyze the data received through the Snapchat application.

Google Cloud is developing a number of cloud services. At the end of August 2018, the company updated the tools for converting speech to text and vice versa. Cloud Text-to-Speech received support for several new languages ​​and voices, and Cloud Speech-to-Text - recognition of several speakers, language and the ability to highlight important words