Forum Replies Created
-
Posts
-
Such a converter would only work on the simplest cases (i.e. nodes connected by network services). FABRIC has a wide variety of dynamically attachable hardware (GPUs, Network Cards, NVMe drives, FPGA), whereas most other testbeds deal in ‘flavors’ of servers that have particular static arrangements of components.
We are pursuing a path of providing notebooks that show how to drive different testbeds via their APIs, rather than the declarative testbed-specific topology representations. RSpec describes only the experiment topology and some experiment configuration and does nothing for the experiment workflow, whereas notebooks that drive APIs do all of three of these things. There is already a notebook example of how to drive Chameleon and FABRIC together. We have an example for CloudLab + FABRIC API example as well that hasn’t been published yet. CloudLab also has a Python API now.
Hello,
As @yoursunny noted please make sure you aren’t actually using the management IP address to communicate between VM slivers as that is discouraged and most of the time not possible.
It is not clear to me if you are using a jupyter notebook or building a slice from the UI (we strongly encourage new users and educators to use notebooks). Please try the ‘Local Ethernet (Layer 2): Create a private local Ethernet network on a FABRIC site (manual)’ notebook (you can find it from start_here notebook under ‘Networking’ section). If it doesn’t work please report which site you are using.
Also please make sure
- You are familiar with this article and the materials pointed out in it https://learn.fabric-testbed.net/knowledge-base/fabric-information-for-instructors/
- Educators are encouraged to post their questions on the Educator Forum rather than a general user forum.
Dear experimenters,
The majority of repairs at StarLight have been performed. FABRIC Dataplane is functioning again, EDC and INDI sites are being released from maintenance. We expect STAR and NCSA to come back as well soon.
Hello,
This error most likely is a result of an outdated fablib version (fabrictestbed_extensions library) in the container. Please
– Remove
fabric_config/requirements.txtfile and restart the container (restart the whole container – File->Hub Control Panel, not the kernel)If the problem persists please report the version of fablib reported by fablib.show_config() call and which container you arae using.
Hello,
This capability is on the roadmap, but not currently implemented. You can share slice access by adding each others’ public SSH keys into the slice, but things like sharing a topology via API is not currently available.
Some workarounds exist like creating a slice in a separate notebook, saving its topology and then using the saved topology (as a file) in another notebook. There are some examples of this in the jupyter examples.
Hello,
We have reopened the original ticket where you requested a teaching project. We will communicate via that ticket (FIP-725)
Some of the equipment connecting STAR site to the FABRIC backbone has sustained damage and is being replaced. We will continue updating this thread as more information becomes available.
In addition to that CloudLab Wisconsin Facility Port is also now disconnected from the rest of FABRIC.
The following sites have been placed into maintenance: STAR, NCSA, INDI, EDC until we know more.
You can test this on one of the IPv4 sites (e.g. I believe MAX or TACC) – if it works there, but not on other sites, the issue is IPv6. You should easily be able to tell if a site is IPv4 by the management IP address you are given when you create a slice (we really should make a list, but it’s pretty short).
If you are trying to reach IPv4 (8.8.8.8) from inside a container, then the issue likely is that you need to enable IPv6 in the container network since the management network you are on is IPv6.
1 user thanked author for this post.
Unfortunately the campus informed us rather abruptly. The slices so long as they are not timing out, should persist – I am not sure about your SSH access into them though. We can try extending them for you assuming they don’t cut our access off completely. Please provide a slice ID or IDs and we can try to extend them.
Hello,
This depends on which site your VM sliver is ending up. It is most likely you are on an IPv6 management site in which case it is worth looking at your DNS setup. By default the DNS servers *should* resolve IPv4 sites using our Nat64 capability, but it sometimes depends on the image.
You can also try manually configuring your DNS to use nat64.net setup.
Read more here about this.
