Closed backeb closed 1 year ago
Hi @backeb ,
How many resources do you need? (more or less)?
Best Marcin
Hi @MZICloudferro
Thanks for your response. I think there might be 10-20 participants of the course... but not 100% sure. Am sending out an email this week to confirm.
How much lead time do you need to prepare the resources? Then I know what registration deadline to set.
Cheers Björn
Hi @backeb
I think we would also need to know if each trainee will require a smaller, dedicated VM or whether they are happy to share the same/bigger ones.
Also, one aspect is providing the quotas and creating the initial VMs, but then the other important aspect is to deploy the required environments on them for the training. Will you need help with that?
Best regards, Sebastian
Hi @sebastian-luna-valero
Thanks for your response!
I think we would also need to know if each trainee will require a smaller, dedicated VM or whether they are happy to share the same/bigger ones.
What would sharing a same/bigger VM look like? Will each user work in their own folder, or will users have to sit together in groups and let one of the users drive?
Also, one aspect is providing the quotas and creating the initial VMs, but then the other important aspect is to deploy the required environments on them for the training. Will you need help with that?
My thinking is to just let the users clone the repo to the VM and then get them to build and run the containers following the instructions, e.g. https://github.com/c-scale-community/use-case-hisea/tree/main/scripts.
@kkoumantaros @MZICloudferro @sebastian-luna-valero is it possible to have singularity installed on the VMs we use? We prefer to run our models through singularity containers.
We have been using docker for the downloading and preprocessing. And if singularty not avail for the model runs can also use docker, but prefer singulariy
Hi @backeb
In principle, having your own VM(s) will allow you to install either docker or singularity.
On the other hand, I had a closer look to the code and here are a few suggestions:
And some questions:
Best regards, Sebastian
Thanks @sebastian-luna-valero
As a user I don't want to install either docker or singularity - I want it to be available for me to use.
Avoid using hard-coded, absolte paths in the scripts to improve portability. For example, this will only work on CentOS but not Ubuntu.
The hard-coded script is just an example of how to run the docker containers. I made a note that users should change the input parameters if they want to use.
It looks like containers are just building conda environments with Python dependencies. Using conda environments directly would remove complexity in the code, and reduce time to recreate the environment for others.
Indeed. Perhaps worthwhile to sit down and discuss how to improve the set up.
My thinking was to keep all the components separate and in separate containers, so that users can pick and choose and use only specific components of the workflow for their own applications.
In my view the workflow solutions are templates that help kick-start a use case.
What's the amount of input/intermediary/output data to work with per workflow run?
Just the download and preprocessing for a 2 day run 4.1G, but most of that data is the global FES2012 data, which is relatively static - i.e. only updated every couple of years or so. FES2014 is available, and FES2022 is coming out soon.
This is excluding the output from a model run and the postprocessing. Am working on running the model.
Will the workflow consume a lot of CPU and/or RAM?
Not sure, but I think its more CPU
How many trainees are expected to join the event?
I expect between 10 and 20, will confirm.
If you want to test conda environments, we can continue the dicussion on https://github.com/c-scale-community/use-case-hisea/pull/34
I am happy to profile the workflow once is ready to better decide how to arrange the computing resources. Maybe, we just need a single big VM where we download input data and provide the conda installation once, and then everybody runs their workflow on their personal area.
Ok thanks re the link to the conda discussion.
I don't want to change my workflow before the training on 15 Sept.
Each participant of the training should
I'd like them to do this so they can get a feeling for how the components of the workflow work. The next ambition is then to string all of these things into a workflow orchestrator e.g. snakemake.
Does that make sense? Maybe we should have a call...
It makes sense, thanks.
Once the workflow is polished, please let me know the estimates for CPU and storage per trainee and I will help commissioning the compute resources in the cloud. Happy to be the guinea pig to profile the workshop as a trainee (and troubleshoot problems). Does it sound like a good plan? Or do you still prefer to have a meeting?
Sounds like a plan!
Next week Thursday and Friday @lorincmeszaros and I will spend the day to try and finalise the last parts, i.e.
It should then be ready for you to test.
The download and preprocess components already work.
@kkoumantaros @MZICloudferro @sebastian-luna-valero is it possible to have singularity installed on the VMs we use? We prefer to run our models through singularity containers.
We have been using docker for the downloading and preprocessing. And if singularty not avail for the model runs can also use docker, but prefer singulariy
@kkoumantaros @MZICloudferro please let me know if singularity is available on your cloud compute.
How many trainees are expected to join the event?
I expect between 10 and 20, will confirm.
Registration deadline for our internal training is 2 September, I will confirm numbers then.
Hi, I came back from vacation, and I am ready to take over this topic on the CloudFerro side. Bjorn, please let me know exactly what kind of resources you need, and I will talk to CF Cloud team.
Łukasz Kubowicz
Hi @LukaszKubowicz
Thanks for your response.
The VM I'm working on to develop this at GRNET has the following specs:
VCPUs | Disk | RAM |
---|---|---|
16 | 40GB | 32GB |
And the VM has docker installed.
The rest can be done in the user space.
With @sebastian-luna-valero we are thinking about one big VM with many users vs many smaller VMs, with one VM per user. I want each participant of the training to do the following:
What do you think, 1 big VM with many users? Or one smaller VM per users?
I will confirm exact number of participants after 2 Sept.
Hi @backeb, I will speak to our cloud team and will let You know.
@backeb @sebastian-luna-valero , I'm waiting for the cloud team to response. But for now on, as long as You have access/ credentials and Project, can You create these VMs on Your own, and manage them ? It might be the most suitable option. Let me know.
Please check properties of these: eo2.3xlarge eo2a.3xlarge
Hi @kkoumantaros and @LukaszKubowicz cc @sebastian-luna-valero, @cchatzikyriakou
After a quick discussion with @sebastian-luna-valero last week, we decided to first talk with @sustr4 about running this training course on CESNET resources. There is already a VO we can use for this purpose.
We want to do this because
The learners will then
I will keep this ticket open, but will include @sustr4 in this conversation going forward.
Cheers Björn
Hi, yes, we are certainly willing to host the training. I did not yet properly read through the discussion above, and it would help me greatly if I could get a summary of the required resources.
Point 2 in Bjorn's comment above is a little confusing to me, but if it means that users are supposed to instantiate VMs themselves, the I propose to use the VO eval.c-scale.eu
to register users. I would make sure adequate resources are available to its members in our cloud.
Keep me in the loop. Zdeněk
Hi @sustr4 and @sebastian-luna-valero cc @cchatzikyriakou, @lorincmeszaros
As mentioned on 📅 15 September 2022 from 15h00-19h00 CET we are running an in-house training course at Deltares where we want to teach the learners to do the following
We have a total of 13 learners No. | Name | |
---|---|---|
1 | Anastasia Zubova | anastasia.zubova@deltares.nl |
2 | Albrecht Weerts | albrecht.weerts@deltares.nl |
3 | Qinghua Ye | Qinghua.ye@deltares.nl |
4 | Sanne Muis | sanne.muis@deltares.nl |
5 | Kun Yan | kun.yan@deltares.nl |
6 | Ruben Dahm | ruben.dahm@deltares.nl |
7 | Bart Grasmeijer | bart.grasmeijer@deltares.nl |
8 | Maarten van Ormondt | maarten.vanormondt@deltares.nl |
9 | Ira Wadani | ira.wardani@deltares.nl |
10 | Marieke Eleveld | marieke.eleveld@deltares.nl |
11 | Rizka Akmalia | rizka.akmalia@deltares.nl |
12 | Bert Jagers | bert.jagers@deltares.nl |
Please let me know next steps.
As for next steps all the people should enrol in the S-CALE Eval VO: https://operations-portal.egi.eu/vo/view/voname/eval.c-scale.eu Enrolment link directly available at: https://perun.egi.eu/egi/registrar/?vo=eval.c-scale.eu
A few additional questions:
How do you plan to access the machines? IPv6, jump host? I can inquire about IPv4 addresses, possibly for a limited time enough could be available.
Will you be needing any specific images, or will you select from vanilla?
As for next steps all the people should enrol in the S-CALE Eval VO: https://operations-portal.egi.eu/vo/view/voname/eval.c-scale.eu Enrolment link directly available at: https://perun.egi.eu/egi/registrar/?vo=eval.c-scale.eu
Ok, thanks will share that info with the participants.
A few additional questions:
- How do you plan to access the machines? IPv6, jump host? I can inquire about IPv4 addresses, possibly for a limited time enough could be available.
I was thinking of accessing the VMs via SSH using a key-pair, which is what we've been doing on GRNET and CloudFerro so far. So for access, lets chose whatever the C-SCALE federation is promoting as best practice.
- Will you be needing any specific images, or will you select from vanilla?
Vanilla? I don't think we need specific VM flavours, we can use what ever is closest to 8 vCPU, 16 GB RAM and 20GB
Before enrolling the VO, please ask trainees to create an EGI account if they haven't done so.
Could you please point them to our docs: https://wiki.c-scale.eu/C-SCALE/c-scale-users/getting-started#create-a-user-account
VMs via SSH using a key-pair
Sure, but that's independent of IP version.
Early warning: trainees should check the output of https://test-ipv6.com/ in order to go with IPv6. For example, my ISP doesn't offer IPv6 connectivity to me (maybe it's only me).
Before enrolling the VO, please ask trainees to create an EGI account if they haven't done so.
Hm, Delaters is listed among Check-in IdPs, but does not actually work. This would have been a nice opportunity to have it fixed.
So for access, lets chose whatever the C-SCALE federation is promoting as best practice.
I´d say the whole Internet community is trying to promote IPv6 as the best practice, but are your attendees capable of using it?
Exactly as @sebastian-luna-valero says. Not everyone can make it work. But if Delatres has IPv6 one premise, that's already a step forward. For all I know, remote workers may be VPNing in, and that would fix all.
I'm currently working from home using a VPN. When I follow @sebastian-luna-valero 's link in an incognito window I get
When I disconnect from the VPN and try again I get
But a while ago I wasn't getting 10/10... so I'm not sure.
Think it might be safest to use IPv4, because that's what we've been using so far...
0/10 at the office. No IPv6 detected at the Deltares office.
OK. I'll see if we still have enough v4 addresses to lend.
Hi @sustr4 and @sebastian-luna-valero
tomorrow @lorincmeszaros and I will be doing some prep work for our training next week.
Could you make sure that a network has been set up in your openstack so that we can do a test deployment of a VM?
The last time I did this, I had some troubles setting up the network (https://github.com/c-scale-community/use-case-hisea/issues/8#issuecomment-897622846). @sebastian-luna-valero indicated that setting up a network would be an action for the provider.
Thanks
Hi,
@sustr4 should I open a GGUS ticket to enable the eval.c-scale.eu
VO at CESNET?
Best regards, Sebastian
@sustr4 should I open a GGUS ticket to enable the
eval.c-scale.eu
VO at CESNET?
Hi, I'll do it locally. I'm on it right now.
Hi again @sustr4 and @sebastian-luna-valero
I've had a look at data resources now that we have most of our workflow figured out.
The storage requirements are roughly:
hydro-nb latest 6199a55e5a06 19 hours ago 5.02GB
postprocess latest c81fc3f8a5ce About a minute ago 1.81GB
download-input latest 0febcb1a315c 7 days ago 2GB
getera latest 239a17fd81cc 3 weeks ago 1.38GB
preprocessing latest a5b1e8fde095 3 weeks ago 1.57GB
continuumio/miniconda3 latest ce7d119281a1 2 months ago 403MB
deltares/delft3dfm latest 80b4d6c3dba6 6 months ago 2.86GB
Total ~15GB
6.5G download
5.9M preprocout
9.0G modeloutput
Total ~16GB
~31GB
So we need VMs with that storage capacity x12 (1 per user)
The standard for all VMs at CESNET is 80 GB. You can set up additional disks or stretch the default ones, but based the numbers I see 80 GB should be adequate.
Brilliant! 80GB is plenty.
@sustr4 do you have a VM flavour in openstack where docker comes preinstalled? or should we install it?
@backeb there is one in AppDB that's already added to the eval.c-scale.eu, so you should have docker pre-installed for you.
Does that mean I can just select that flavour from openstack when I instantiate a VM?
Yes, that's the goal.
Does that mean I can just select that flavour from openstack when I instantiate a VM?
Let me know how that goes :-)
Hi @kkoumantaros @MZICloudferro @LukaszKubowicz cc @sebastian-luna-valero @cchatzikyriakou @lorincmeszaros
At the EGI conference we are planning a demo and training of the hisea use case. In preparation for this demo / training we are organising an internal training at Deltares on 15 Sep from 15h00 - 19h00.
@kkoumantaros @MZICloudferro @LukaszKubowicz would it be possible to set up VMs for our participants to use during this training?
Thanks Björn