[Ilya Baldin]

Susmit,

For the moment what you describe is the way to do it. We are considering a feature to allow users within a project to see each others’ slices, but it has not been implemented. One thing that can make your life a little easier is the ability to introduce other users’ keys into your slice at creation time described in this notebook:

– https://github.com/fabric-testbed/jupyter-examples/blob/b9019e5a7b3b4eea26695fab2257d44b691b168e//fabric_examples/fablib_api/other_ssh_keys/add_keys_into_slice.ipynb

 

[Ilya Baldin]

Hello,

FPGAs are still largely in testing and not everything works yet. For example DPDK cannot yet be used with FABRIC FPGAs. Also FABRIC FPGAs are there specifically for *deploying* bytecode developed *elsewhere* (in your lab or in another testbed) into a large topology. FABRIC is poorly suited for initial development of the bytecode because FPGAs are passed to VMs and initial development frequently requires cold reboots of the underlying server which affects other experiments running VMs on the same server. You can tweak existing code on FABRIC (because that generally just requires a VM reboot when the FPGA is updated), but developing new code from scratch is not feasible.

So if you

a) Understand the constraints I described above

b) Have already developed an initial version of the code somewhere else

you can fill out a project permission request from the Portal and request FPGA permission. Please specify which shell you are planning to use (ONS or XRT).

Other testbeds (Chameleon, OCT and CloudLab) have similar U280 FPGA resources attached to bare-metal servers where the initial development should take place.

[Ilya Baldin]

Hello,

No the largest RAM allocation for a VM in FABRIC is 384GB as described in this article.  However other testbeds connected to FABRIC have large memory machines. For example Chameleon. Your experiment may be better suited for Chameleon, or, perhaps, a combination of FABRIC and Chameleon using the Facility Ports that exist between two testbeds. There is a notebook that shows how to construct an experiment between the two.

[Ilya Baldin]

Hello,

I’ve added you to the tutorial project. Be sure to select that project when creating slices.

For your class – in the questionnaire you did not indicate any need for specialized resources. FABRIC has a relatively small number of these (GPUs, SmartNICs etc) so we prioritize their use for researchers over educational uses. If you think your class will require their use you will need to fill out the class request form again with an updated description so it can be reviewed.

[Ilya Baldin]

Most specialized components require a permission on the project to use them (please take a look at the linked documentation). The project that was created for your class does not have those permissions because you did not indicate the need for that. If you are simply trying to learn about FABRIC and want to use some specialized components I can temporarily add you to the tutorial project that has many specialized permissions – we can keep you in that project for a few weeks.

[Ilya Baldin]

This is being answered in another forum.

[Ilya Baldin]

Yuvaraj,

In general we very strongly recommend that new users start by using our Jupyter Example notebooks. Please take a look at the Getting Started section of the documentation and from there try the notebooks.

[Ilya Baldin]

Fraida,

You are correct, because we are building the testbed in phases and each site has its own image store, we so far have not been wholly diligent about making all kernels exactly the same. We will put it on the list to fix once the the building rush is over.

[Ilya Baldin]

Thank you for the suggestion. Let us look into it.

[Ilya Baldin]

It will not. It will do a diff between the existing topology and the one you want and make the necessary changes. In some cases it may reboot the existing VMs.

[Ilya Baldin]

I cannot answer readily what the problem in your setup is.

Tuning network performance to 100Gbps and beyond depends on a large number of parameters. The notebooks Paul pointed out show some of them. Things like

– number of available cores and RAM

– affinity between vCPUs, physical CPUs and network cards

– type of network card (to get close to 100Gbps you need dedicated not shared NICs)

– MTU size

– Number of threads used by the data transfer app

– etc etc etc

Network service type will not affect the performance. The type of network card used will.

[Ilya Baldin]

I updated the section. For me at least regeneration has worked in the past, but Stop/Start is probably more reliable, although more disruptive since you have to pull up all the tabs with your notebooks again.

[Ilya Baldin]

Please tell us which version of fablib you are using (it is displayed in the notebook as part of the first fablib.show_config() command and also which version of Jupyter Notebooks you are using.

Also please indicate if you are using our Jupyter Hub or trying to run from your laptop.

Your error indicates that the user you are logging in as is not in docker group and cannot start Docker containers. You can remedy this by doing

$ sudo usermod -G docker rocky

However this **should not be necessary** if you are running a recent version of everything.

[Ilya Baldin]

UMich campus network is reachable again, we are cleaning up the MICH site and preparing to take it out of maintenance.

[Ilya Baldin]

In general though – this is what we recommend – go to Hub Control and restart your server – that reinitializes everything properly. I’m not sure what machinations Jupyter Hub does when you simply do ‘Restart’ but it is different from going into the panel and doing Stop/Start (or sometimes just Start, because it detects the server has stopped).

[Author]

Posts