In general, the bottleneck is always going to be the
N_students*resources_per_student.

That varies widely by what the students are doing. In our data science
courses, we give each students 2GB RAM and approximately 4 students
per CPU core. The CPU stuff isn't typically the issue - RAM *always*
is.

For more basic python programming stuff, you could use less RAM per
student, but I wouldn't go below around 500MB per student.

Cheers,

Brian

On Thu, Feb 16, 2017 at 3:53 AM, Doug Blank <[email protected]> wrote:
> On Thu, Feb 16, 2017 at 6:23 AM, Thomas Kluyver <[email protected]> wrote:
>>
>> On 16 February 2017 at 10:00, DVD PS <[email protected]> wrote:
>>>
>>> we won't deploy it till we have some understanding of how many we can
>>> handle
>>
>>
>> I don't know of any tools to help you with this, unfortunately.
>>
>> Running the notebook servers takes ~40MB memory per user, and each Python
>> kernel started is another ~30MB before loading any libraries. But resource
>> use will often be dominated by what code users are running inside their
>> notebooks.
>
>
> We've been using JupyterHub since it was almost ready to use, and there
> isn't an easy way to answer the question "how many students can a machine
> handle?" We use a variety of kernels, and some languages are better than
> others. We have some Javascript kernels that have a small impact on the
> server, and Java kernels that just a dozen of students can bring a machine
> to its knees (at least for a short time, while compiling). IPython is pretty
> good, but of course it depends on what they are doing.
>
> We use JupyterHub across Physics, Biology, and Computer Science. The Physics
> courses tend to have the highest load per cell, partly due to their style of
> processing, but also because of their problem sets.
>
> Our server has 512 gigabytes of RAM, and 12 cores. It is actually just the
> head node on a cluster. Our goal is to get it set up such that a student's
> kernel would spin up on the node that is most available (probably using
> docker). But we have to figure out how that plays with the other jobs and
> the scheduler.
>
> I had tried a CPU limiter in the past. Our load is getting so high now, we
> may have to revisit that.
>
> I'd be glad to help answer this question as we have a live, operating setup.
> Feel free to contact me directly if there is something I can do (e.g.,
> provide other stats, run a test program, etc.)
>
> -Doug
>
>>
>>
>> Thomas
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "Project Jupyter" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to [email protected].
>> To post to this group, send email to [email protected].
>> To view this discussion on the web visit
>> https://groups.google.com/d/msgid/jupyter/CAOvn4qgxVXwS97MUQXxRQ8FioJh_DmjOTtcX9-KottgLmkVfJw%40mail.gmail.com.
>>
>> For more options, visit https://groups.google.com/d/optout.
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "Project Jupyter" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To post to this group, send email to [email protected].
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/jupyter/CAAusYCgQ4LWvd-btUC8z%3DgbR5u6q2GWsY%3DwcBuJVoEOEwgMwdA%40mail.gmail.com.
>
> For more options, visit https://groups.google.com/d/optout.



-- 
Brian E. Granger
Associate Professor of Physics and Data Science
Cal Poly State University, San Luis Obispo
@ellisonbg on Twitter and GitHub
[email protected] and [email protected]

-- 
You received this message because you are subscribed to the Google Groups 
"Project Jupyter" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/jupyter/CAH4pYpSNRL7miCwDCLd%3D%2BJLRCWW3RbS%2B-r%2BgeGa7QOmTwVds4Q%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to