Congratulations to @rstutsman for winning the #UofU College of Engineering Outstanding Teaching Award, and to @carboxylman (who works on #CloudLab, #PowderWireless, and other projects) for a CoE Outstanding Staff Award!

@shefys Aw man, I remember back to when someone from Arm told us that we might have the largest Arm cluster in the world in #CloudLab with 315 machines. That was only about eight years ago, how things have changed.

@elb I actually use scribble quite a bit, all of the #CloudLab and #PowderWireless docs are in scribble. And I use Butterick's pollen for personal website. I thought he was doing something print-focused years ago, but when I looked it seemed to be dead - I'll have to look again!

I'm super proud of my student @lremes who just defended his Bachelor's Thesis. It's a distributed #fuzzer, Hopper, and he got it running with up to 240 distributed process on #CloudLab. It's been great working with him for the last few years.

We'll get his thesis posted soon, but Hopper is already up on github: https://github.com/Cybergenik/hopper

Last week, I ran a panel with @dave_andersen and Cody Cutler covering the history of the Flux Research Group at the #UofU and testbeds like #Emulab, #CloudLab, and #PowderWireless . The video is now up: https://www.youtube.com/live/LUFp6sjKbkE?feature=share&t=15360

(This is a particular timestamp in a video that covers the whole day of talks about the 50-year history of the Kahlert School of Computing and by various alumni.)

I took some folks on a tour of the #UofU downtown #datacenter, which houses part of #CloudLab (along with a great many other things.)

Pictured: Aleks, one of the primary CloudLab administrators and the person who works most directly with the hardware. Hannah, who has been doing #HCI studies of how people use CloudLab (thanks to those who have participated!); Pavani, who is working on fine-grained measurements of latency and building a pseudo-layer1 switch using #P4; and Sachin, who is working on the #security of #ssh and network security processes more broadly.

Fuzzing readelf on a 48 core machine with my #distributed fuzzer Hopper:

https://github.com/Cybergenik/hopper

I have 4 more of these doing the same thing, lets see if I find something interesting.
#cloudlab #fuzzers #gnu

@dwallach As someone who has, regrettably, designed a lot of tooling around XML (all of #CloudLab runs on an XML intermediate representation of hardware and topologies), I was unfortunately late to reach the conclusion that XML is a write-only format

@infinimatt @dan @kstatz12 We promise that #CloudLab provides as many 8s of uptime as you want.

Yesterday we took a group of folks from industry, government and the press on a tour of some of the #PowderWireless infrastructure.

This is one of five cell-like towers we own on the #UofU campus. (This one is on Wasatch Drive near the roundabout, for those familiar with the area.) These (plus one more on a building) are placed so that they cover one of our campus bus routes; the buses have end-user-type devices on them that can connect to these towers.

The cabinet at the bottom hosts a collection of software-defined radios that #wireless #research folks use to build their own customized #5G networks - or, because they are #SDR, really, anything, including #spectrum monitors, #LoRA networks, channel sounding, etc.

There is some compute available right there roadside, plus these are connected by fiber to some big datacenters nearby including #Emulab and #CloudLab.

https://powderwireless.net/dense

@dev @secretasianman I have had a NSF program manager complain to me that se got support letters from so many #CloudLab users that they had trouble finding non-conflicted people to be on a panel, and I consider this as an absolute win.

@lschuermann @kalyanaj Yeah., I still need to fully embrace the #nix way of doing my own development; the few times I've tried to convert some of my own projects to building them as nix flakes, I've run into roadblocks: for example, stuff that wants to write files in its own source directory, or #racket packages that don't seem to be as integrated into the ecosystem as, say, python ones. They're solvable problems but so far I just haven't put in the time.

Really, one of the things I want to do with #NixOS is to get first-class support into #CloudLab for it, with the goal of giving people another powerful tool for reproducible experiments there. It's a pretty good fit, I think, as the overall workflow there is to describe a thing as a profile and treat the artifact itself as a sort of a temporary build artifact. With images described in the nix language rather than by building them imperatively, you could really capture pretty much an entire network environment in one place, reproduciblyl

@lschuermann Sure, #CloudLab is a service we really enjoy providing!

A tip for #CloudLab users:

You probably already know that most CloudLab profiles are defined using python scripts - when you instantiate the profile, we execute the script, filling in any parameters you provided, and the output describes the nodes you're requesting, which disk images to load, etc. See https://docs.cloudlab.us/geni-lib/index.html for examples.

But: did you know that you can install geni-lib, the library that does the heavy lifting in this process, locally to debug your profile scripts? Grab it from our gitlab at https://gitlab.flux.utah.edu/emulab/geni-lib and you can execute profile scripts on your own machine, examining the output to make sure it's asking for what you intended. It even has the ability to interact with our API directly, allowing you to script creation and teardown of experiments.

@ricci #cloudlab has been amazing to use and run experiments! Couldn't thank you folks enough for all the work you put in to make sure cloudlab is up and running always!

Early to mid December is the busiest time on the #CloudLab testbed. The two biggest factors are the #OSDI paper submission deadline and the end-of semester crunch in classes. Our users are running 609 simultaneous experiments right now!

These graphs show node availability, starting today and going out through the next several days. As you can see, almost all node types have zero or just a handful of nodes available today[*].

To help alleviate this, we've built a reservation system that allows users to request time on nodes in the future - the points on the graph past today give you a sense of what's available to be requested for reservations.

Good luck with your papers and courses, everyone!

[*] There are a lot of m400 nodes available on the Utah cluster - these are server-class ARM machines, but are several years old, hence the lower interest in them.

@vish FWIW, this is caused when all nodes are allocated or reserved, and then some fail. Makes a mess for everyone

#CloudLab

Something I've been experimenting with in the last week for #CloudLab. This is a custom carrier board we built for @Raspberry_Pi CM4s that provides dual gigabit networks (w/ onboard switches to save on wiring) and remote power cycling, serial console, and imaging. Once we can work out a way to get around supply chain issues, we hope to add dozens of these boards (with hundreds of CM4s) to CloudLab for research and education on #IoT and #edgecomputing.

#raspberrypi #cluster

If you have used #CloudLab - even just a little bit - consider participating in our #HCI study regarding how people use it and what we can do to make it better.

All sessions will be conducted online and will not take more than 60 minute. We will ask you to log in and show researchers how you use #CloudLab, then ask you to participate in a short interview.

Visit https://cloudlab.us/study if you're interested in participating. For more information or questions, contact hannah.cohoon@utah.edu .