SNU Data Science Seminar - Spring 2021 - #4 - Shared screen with speaker view
Who can see your viewing activity?
장창섭 / 학생 / 데이터사이언스학과
Thank you for the talk today. When using docker for data science, would data transmitted among data scientists / owners be encrypted (and analyzed without disclosing original data)? What is your opinion on the data privacy in cloud setting and how docker is coping with privacy concerns around data?
When data scientists work, they doing a lot in notebook environment (e.g. jupyter). Therefore, it is inevitable to "containerize" code written in notebook environments to deploy it to production, and it's overhead. How will these problems be solved in the future?
Thanks for the talk! I am not really familiar with docker, but I think docker is somewhat similar to kubernetes or Anaconda. What are the strength of docker over docker's direct competitors?
According to what I know, for web scale, I have to open a port for docker container to act as my backend. How can I assure the security for that container port?
I understand that docker is powerful on data science. But is there any other use-cases? especially some systems that is between hardware and software such as robot control system, etc ?
What are the advantages of Docker for HW developers? Let's say I develop a new neural network accelerator on an FPGA and want to distribute my work to many communities and researchers.They need applications and also device drivers or other necessary Linux kernel modules to reproduce my work on their servers. How does Docker handle this situation?
Comparing with 2016 when I had used docker first, docker hub and other communities are very popular and developed. But there are some issues using big hardwares such as terabyte-scale main-memory servers, and using huge volume of dataset, there was some configuration issues such as volume manager, NAS connections.
There were many solutions on google but it's not qualified and it didn't fit our demand. To configure small things makes harder to use docker to make our experimental environment. Any solution to qualify shared containers or to make easier exploiting hardwares in many environments?
nvidia-docker and docker-cli are separated packages as I known. How can I integrated of those when I build my docker on a local machine without GPU and deploy in a server with GPU?
It seems that all the container images commited to docker hub have private data. Also, for the security reason, all the images commited to docker hub should be encrypted. It means that all the data are hardly reducted using data reduction techniques such as deduplication. How does Docker handle the huge size of data in docker hub?
what is your suggestion for engineering students to run a successful company like Docker? Should we apply for an MBA program?
Great thanks. It was very insightful!
Hwang Ye Jin
편명장 (Myeongjang Pyeon)
Thank you :)