Skip to content

Home

Jump to bottom Edit New page
Archit Sharma edited this page Apr 12, 2015 · 10 revisions

FAQ

So what is dockerComp again? how do you achieve better distributed computing?

Refer to Introduction in https://github.com/arcolife/dockerComp/blob/master/README.md Basically, its an energy efficient platform to perform distributed computing, by harnessing power of idle CPUs across the "online" machines in this world, distributing workloads, get processed data back and perform analysis. I've introduced dummy data for prototype right now. But when the workloads/actual agents are added, we hope to achieve better performance than VMs as theory shall indicate.

Jiannan (https://github.com/zhangjiannan), who is helping me out with this project, says- "DockerComp is definitely related to energy saving, because large-scale volunteer computing will save a lot of energy from data centers". And I concur!

Whats to be done to enhance the prototype?

I didn't have the agent that runs inside VM, any dummy workload you may put, that distriubutes data from server to the clients and further to N number of containers inside each client and then they use the client's CPU to analyse and return results.

Is there anybody doing that already?

No one as far as I know. Docker is new, atleast in scientific community, as per the best of my knowledge. [google searches and blog reads ( :D )]. Everything currently runs directly on servers or in VMS. But what does it matter? Do it better in this manner. #UNIX philosophy.

Don't worry. Facebook/Google wouldn't have existed if they weren't better than the rest. Just saying!

What's more in the closet?

Imagine if that crowd sourced analysis could be done efficiently, in lesser time and also, being able to analyze more data in that time being able to run multiple agents (within multiple containers) on multiple clients and being able to control the number of containers launched per client that would increase even the chances of finding aliens :D through the SETI project that uses crowd sourcing. They all run things like BOINC and stuff that ultimately runs inside a VM.

I really wish to take this further and finish this as a pluggable dockerized generalized distributed computing framework

So where does the app run ultimately?

It runs inside Linux machines (Ubuntu/Fedora) ..i.e., RPM or DEB based systems. It currently doesn't run on Mac. We need to add suport for that.

I made a basic prototype that launches certain number of containers and makes it easy and simple for users by having to run just one script at the end install_me.sh. No need to download images for VMs everytime, have centralized git-like docker-hub based image control and distribution system, push changes as stacked images and no need to download oracle virtual boxes... nothing, just one simple script.. install_me.sh.

Something you should know..

I was facing problem with running the app on client exists in network proxy. In that case, it's generally tricky to initiate request from outside, especially if its a PAN/NAT network type (as per my knowledge). So, I sought to initiate the request from client and developed a method to keep it alive and let the server communicate with client in that manner. The server IP is hardcoded inside client's test script for checking connection to server. (and this could be controlled from the images distributed from central repository). Needs to be more sophisticated though.

..but usually servers should initiate the connection, right?

Right. need to work on that.. tunneling I guess.. I didn't have time and resources for that. Contributions are welcome!

Task Work-flow

If there are different tasks, there will be a task ID thats tracked on server inside MongoDB. Right now i'm launching a fixed no. of containers on a client; like 4 docker containers. This needs to depend on the h/w config on the client otherwise it will hang the CPU

TODO

make a daemon on client side, that keeps track of containers and allots CPU power and also communicates metadata with server.

TESTS

  • From client side:

    • although the default connection establishment test is included with install scripts; run $ ./client-side/test.sh

  • From server side:

    • TBD

  • Workloads:

    • Currently a simple task. TBD.

WORKFLOW

  1. Server

    • Dashboard to Manage:

      • No. of Clients (and # of containers per client)
      • Resources allocated to the containers

    • Master app that manages data sent to each client and checks for integrity.

  2. Client

    • Installation of Docker
    • Starting Containers
    • Installation of Application inside the Container
    • Connection Establishment with the Server.
    • Scripts for the computation
    • Error Reporting

REFERENCES

  1. https://github.com/cernvm
  2. http://en.wikipedia.org/wiki/List_of_distributed_computing_projects
  3. http://www.rightscale.com/blog/sites/default/files/docker-containers-vms.png
  4. http://www.psc.edu/science/
  5. http://pybossa.com/
  6. https://okfn.org/press/releases/crowdcrafting-putting-citizens-control-citizen-science/
  7. http://www.mediaagility.com/2014/docker-the-next-big-thing-on-cloud/
  8. http://cernvm.cern.ch/portal/
Add a custom footer
Add a custom sidebar
Clone this wiki locally