... and the bloody thing is still running after 22 years.

 This is disgusting. Notice the kernel version 😎 #linux.

Can you make a useful student terminal using a Raspberry Pi 5 ?

---

Executive summary

Yes, you can. And not only it is doable, but it is actually good fun! The Pi’s are fully useable as student terminals, allowing students to do much more than just ssh-ing to the servers : they have a fast and responsive local environment with chromium, libreoffice, molecular graphics, any programming language they want, R & R-studio, etc. Having said that, the students using them must not be in fear of doing some system administration now and then (but with a carefully prepared image for the microSD card, even this will be very rare). All-in-all, this is a loud and clear go from my side (and at the present time).

The Pi at full throttle. Overview of four workspaces shown with
molecular graphics, libreoffice writer, a tmux session, and a browser. 

---

The long story

Introduction

So, I decided to buy a Raspberry Pi 5. Not sure why. Probably to have a new toy to play with during the Christmas vacation. This turned-out to be a good choice: the Pi is very good fun, and I still enjoy fiddling with its settings and making it look like a proper linux machine. There were some minor issues, like a difficulty in finding suitable adaptor for micro HDMI, or -the several- issues with Wayland/Xwayland (more on that later), but the final verdict is very positive : Pi is nice and useful as a desktop environment, the hardware looks stable and the software dependable (but it is too early to say), and at 3W power consumption at idle (and 12W under full load), it will make the university happier. Actually it is so nice and the desktop experience so smooth and responsive, that I am seriously thinking about moving it to my desk at the office.

Hardware things

Not much to say here. The things I should mention are : (a) The Pi’s micro HDMI port is a bit difficult to work with, and you have to think whether you want to buy an adaptor or a fixed cable. (b) The official Pi.com case is not very good at cooling the Pi, maybe the official ‘active cooler’ is better and cheaper. There are also these ‘cases’ with double heat sinks and double fans for the fanatics. (c) It would be nice if your monitor comes with speakers, otherwise you will need either a USB- or a bluetooth-based set of speakers/headset/earbuds. (d) The ‘disk’ is a microSD card. My understanding is that this is the weak link hardware-wise. MicroSDs are not made for 24/7, you probably need a high-end model and properly scheduled backups to be on the safe side.

Software things

The Pi comes with Debian 12 (Bookworm), so everything works out of the box, except Wayland/Wayfire. I suffered considerably with the move away from X11, but in retrospect most of it was due to the complete absence of any familiarity from my side with the configuration of wayland and friends. Eventually we did reach an equilibrium and I am happy (for now ?), but note that some things like camera support and video-conferencing have not been tested. The Xwayland part (offering support for pure X11, non-wayland, apps) is not exactly a true substitute for a proper X11 server, and I even ended-up re-writing some code for two of my ancient programs which were using Ygl (an IrisGL-like X11-based substitute for GL) to make them run as fast as they do under X11. Having said all that, I do not regret staying with wayland : I learned things, and wayland appears to be faster and smoother and ‘fancier’ that X11. There are things that bug me, like the absence of a focus-follows-mouse setting (you need to find/build the extra plugins for wayfire to make this work), but all-in-all it looks good now. I suppose that someone with fewer preconceived ideas and expectations could have used the defaults happily and never experience any issues, but this was definitely not the case with me.

Tests in a ‘production’ environment

The Pi passed what I consider to be an acid test for a new desktop machine : log-in onto it, and try to work normally while using it for dealing with whatever happens to come along during a typical day at the office. I have been happily using it for several days now, and only occasionally did something needed to be installed/configured. The general feeling I got is that of using a normal fully-setup modern linux machine on a hardware that is up to the task.

A workspace with too many windows open.
This is from an analysis performed remotely via ssh.

Cost estimates

It looks like you are better off buying a kit and then add to it if needed, instead of buying the parts one-by-one. For example : buying separately the 8G model + the official Pi case + the official power supply costs 123.90 euros (incl. VAT). Buying this kit costs 124.90 euros, and you get (for the same price) the 8G model + the official active cooler + the official power supply + a HDMI ⇒ MicroHDMI cable + the (cheap-ish) ABS case + a Sandisk Ultra 32GB MicroSD (with pre-installed the OS). If 32G is not enough, you only have to buy a larger/better MicroSD card, attach a keyboard and a mouse, and you are good to go.

So, even if I add 30 euros for an extra MicroSD card + a set of wireless keyboard/mouse, we reach a total of ~155 euros for a complete ready-to-go Pi box. Add ~90 euros for a cheap 23" monitor with speakers, and you end-up with a grant total of ~250 euros for a complete Pi-based student terminal. Now : For 190 euros, you can buy a cheap chromebook with 11" display, two cores, and a 32G disk. But I can hardly imagine running e.g. pymol on a 11" display. I am writing this on a chromebook in vim and it is fine for that usage. But the chromebook is not a suitable device for molecular graphics, or having 5 windows open from an analysis of a molecular dynamics trajectory. To conclude this section : if you already have the monitor plus keyboard/mouse —and the 32G MicroSD is acceptable— you can buy a proper desktop machine with 125 euros, which is probably unbeatable. Even if you have to go for the full set, I doubt that you could find an as smooth and responsive machine with 250 euros.

Conclusions

The Raspberry Pi 5 is not a toy-like hardware suitable only for kids learning computing. It has sufficient resources to make a useful everyday terminal for college students as well. The Pi is ideal for writing code, testing it, and then deploying it to the servers for the production run. With the 8G model, you can easily have 4 workspaces loaded with a browser, a local tmux session, a remote session with X11 apps running an analysis, a local diary/journal for taking notes, etc., and with all those apps running, the responsiveness of the Raspberry remains essentially unaffected. You will, of course, not do your heavy computing on the Pi, but this is also true for all terminals we currently use. And with a price tag of 97€ for the 8G model, replacing a broken terminal comes as close to a walk in the park as you’ll ever get. Add the 3W power consumption at idle, and there you have it : a functional solution for light academic computing that is also light on the budget and also light with its environmental footprint. The way I see it, the only thing remaining to be seen concerns the robustness of the Pi : our classical PC-based student terminals can easily pass the 5 year mark, and we have several that have passed the 10 year mark. For the Pi to be useful, it should at least approach the 5 year mark. Replacing MicroSD cards every couple of months would definitely take the fun out of it. Time will show …

Update [April 2025] : Not only the Pi is still my main desktop machine (and I'm happy with it), but we have actually moved ahead and installed three devices for the students' office. The photo says it all :

A fully useable academic laptop for 189 € ? My take on a dirt-cheap chromebook.

 

I decided to write a review about this device¹ mainly because it pleasantly surprised me. My 5-year-old linux laptop took a hard fall, the casing broke, and I felt that to be on the safe side, I needed a ready-to-go backup device for this semester's teaching classes. I basically needed a device with a useable unix shell, an editor, gcc, a pdf viewer, and to be able to connect to the departmental projectors without issues. I stumbled across this chromebook by chance (actually, by sorting on the laptops' prices), and it caught my eye.  I was always curious about what google is doing with chromeos and chromebooks, but my expectations from a 189€ chromebook with a weak CPU, 4G of physical memory, and only 32G disk space, were low.  Extremely low. And I was wrong. This is a fully useable laptop for normal everyday academic applications. Having 4 workspaces ("desks") with dedicated email, unix shell, browser, a video call, and with all apps running simultaneously and being responsive and useable is the norm and not the exception. Add the ~10 hours of battery life on a single charge, the immediate availability of a proper linux (debian) distribution, the appealingly minimal, functional and stable chromeos environment, and it doesn't take long before you start thinking about taking it with you every time you leave home. And with a weight of 1.1 kgr, you might as well take it.

Don't get me wrong : you won't load this device with your brand new 10 Gbytes data set, and you will certainly not analyze the data with its AMD A6 processor. And you almost certainly won't do any short of serious photo/video editing or gaming on this chromebook either. But you can easily do all teaching presentations, all emailing, all everyday administrative tasks, right from this device.

However, do bear in mind that I'm an old guy. So old, that the following caveats apply :

▶ All my important/needed files fit in 3 Gbytes. If you need 64 Gbytes just for your 'must have' files, you need a bigger machine.

▶ I do not mind working with google docs, and storing files on google drive. Having said that, I also have my dropbox files on the device using the 10 Gbytes linux partition (which is shared with chromeos). If you avoid the cloud and want everything stored locally, or if you definitely need the microsoft office environment, chromeos may not be for you. Note, however, that you can have libreoffice via linux if you prefer local storage.

▶ I do most of my stuff through the unix shell : email with mutt, editing with vim, reading/annotating PDFs with llpp and evince, image viewing/editing with eog and gimp, keeping my lab book with rednotebook, etc. I do use google docs for the occasional docx and xlsx, but normally I prefer typing in the unix shell. The implication is that neither unix, nor me, are heavy on the resources required. If your applications or files are heavy, you may need a bigger machine.

▶ I liked chromeos within hours of using it : it is minimal, functional, non-intrusive, and very pleasant to the eye. But if you prefer the more heavy and rich (in distractions ?) environment of windows and macos, then you will soon be bored to death with chromeos.

▶ One thing I definitely didn't like was the end-of-updates date for this device. If you buy a model made after 2021, you are getting 10 years of support for chromeos. But this device was made in 2020, and its support ends in 2026, which is too soon for a device bought in 2023.

Allow me to close this review with the things I liked most about this device. The first was that with no effort from my side, I got a 'user-friendly/GUI-based' operating system (chromeos) together with a proper linux distribution (complete with all debian packages, including those with a GUI). No more fighting with bootable USB sticks, or with the window's WSL2. One click to enable the linux environment, and you are good to go. The second thing that I liked a lot is that chromeos  appears to be bloat-less and minimal by design (if not by force). You start the device, and not much is happening. No by default pop-ups, no notifications, no beeps, no nothing. An invitation to open your apps and do whatever you wanted to do in the first place. Last -but not least- is, of course, the price : a brand new laptop for 189€ is probably something worth writing home about.

All in all, these were 189€ well spent.

--

¹ The device : Lenovo IdeaPad 3 CB 11AST5 11.6", A6-Series-9220C/4GB/32GB.  Full specifications from here.

² Note added in proof : chromeos is indeed very light. I installed the ChromeOS FLEX edition on an ancient PC dated back to 2008 (Acer Veriton X270 SFF Intel® Pentium® E5300 4 GB DDR2), and the bloody thing is responsive. Impressive given that this is 15 years old hardware (which was running ubuntu 10.04).

 A new fat node has been added to the cluster. This is based on an Intel Core i7-13700KF (1700/3.4 GHz/30 MB), an NVIDIA MSI RTX 4070 Ti VENTUS 3X 12Gb OC 3, 8GB DDR4@3200 MHz, and two 2TB disks in a RAID1 array.

Initial tests confirmed that this is our new fastest machine, delivering with gromacs and HMR ~3930 ns/day on the FipWW 8,000 atom/5fs system. As with the other machines, with smaller systems the performance flattens at approximately ~4.1 μs/day (again with gromacs & HMR).

New node added to the cluster

A new fat node has been added to the cluster. This is based on an Intel i9-9900K, an RTX2070S, 8GB DDR4@3000 MHz, and a 2TB disk.

Initial tests confirmed that this is our new fastest machine, delivering ~575 ns/day on the standard 10,000 atom/4fs benchmark. As with the other i9, with smaller systems the performance flattens at approximately ~700ns/day with a 2.5fs step.

New fat box arrived

A new fat node has been added to the cluster. This is based on an Intel i9-9900K, a GTX-1080, 8GB DDR4@3000 MHz, and a two-disk RAID 0 @ 4TB.

Initial tests confirmed that this is indeed our new fastest machine, delivering ~560 ns/day on the standard 10,000 atom benchmark (second best is this one with 460 ns/day). With smaller systems, however, the performance flattens (giving with the CLN025 system 570 ns/day@2fs, ~730 ns/day@2.5fs).

Running dropbox on old linux machines

Dropbox recently changed the linux system requirements. This may be a meaningful choice from the company's perspective, but I am not prepared to give-up on some of my old machines which have software that I find it hard to move to newer hardware. So, I was looking for a software solution that could work with ubuntu 10.04, ubuntu 11.04 and a relatively new centos machine which uses xfs instead of ext4. First I will describe the technical steps for the procedure I'm currently using, then I will discuss the shortcomings.

Technical steps (based on rclone and inotifywait)

[1] Install rclone as described in https://rclone.org/downloads/ . Surpisingly, this may be the hardest part of the procedure due to old versions of curl not being compatible with the server-side ssl. On both of my machines I had to install openssl from source and then install curl again from source (using the newest ssl libraries). Expect suffering at this stage.

[2] Configure rclone as described in https://rclone.org/dropbox/ . This is easy.

[3] Do a dry run to confirm that rclone is working as expected (change 'mydropbox' and 'user'):

rclone --dry-run sync -u mydropbox: /home/user/Dropbox

[4] Install inotify-tools if you don't have them already.

[5] If you reached this stage, you are practically done :


(a) Prepare a hidden file in your home directory (for example /home/user/.dropboxwatch.sh) containing the following bash script :

#!/bin/bash
while true; do

inotifywait -e modify,create -r /home/user/Dropbox && \
 pgrep rclone > /dev/null || rclone copy -u /home/user/Dropbox mydropbox:

done
exit

Make this file executable (with chmod 755 /home/user/.dropboxwatch.sh).

(b) Edit your crontab and insert the following two lines :

@reboot      /home/user/.dropboxwatch.sh >& /dev/null
*/30 * * * * pgrep rclone > /dev/null || rclone sync -u mydropbox: /home/user/Dropbox

(c) Add the following to your .bashrc :

alias dropboxfetch='pgrep rclone > /dev/null || rclone copy -u mydropbox: /home/user/Dropbox'

(d) Reboot the machine and confirm that .dropboxwatch.sh is running.

What you get from all that

(a) When you add a file to the dropbox directory of the local (old) machine, it will be immediately transferred to the dropbox servers (and, thus, to all of your other machines).

(b) When you modify the dropbox directory from other (non-old) machines, the changes in the old machines will appear at the time interval defined in the crontab (every 30 minutes in the example above). If you want to immediately fetch new files from dropbox to the local (old) machines use the dropboxfetch alias.

(c) You can not push file deletions from the old machines to dropbox. Dropbox files can only be deleted from newer machines that run the real dropbox daemon. The deletions will appear at the crontab specified interval. Think what you may, but this is a great precaution (and do note that at no point do we do a sync from the old machine to dropbox, which is the definition of better-safe-than-sorry).

Things can definitely go wrong. Notice for example these pgrep's ? These are to avoid some obvious race conditions that arise if you end-up having two rclone jobs running simultaneously. But I bet (especially with dropboxfetch) you can still get artifacts. The thing is that in the absence of a sync from the old machine to dropbox, the worst case scenario is that some files you thought you deleted, may end-up being well and alive. (famous last words I hear you say).

Gemini PDA review

I received the Gemini seven weeks ago, so this is possibly a bit too early for a review, but there you go. To my mind the basic question that has to be answered is this : If I could turn the clock back, would I still choose to buy this device? And the second (Gemini-specific) question is this: What is the Gemini ? Is it a phone with a physical keyboard or is it a mini laptop that can take phone calls ?

My answer to the first question is an unequivocal 'Yes, I would buy this device again'. This is the first device -in almost two decades of carrying with me a phone- that feels more like a proper computing machine (that I can do productive work with) than yet another smartphone. And yes, the previous statement about computing machines does gives me away : I humbly admit that I really do appreciate the ability to open a unix shell through termux and connect via ssh to the servers 900km away, check the jobs are running, plot a graph (through ssh/xsdl), edit a file with vim, save it, restart a job, logout and all that with all the special control characters readily and easily available. I do appreciate that I can properly compose an answer to an email without the distraction of fighting with the on-screen keyboard (and even send it properly signed with S/MIME). I do like the ability to open a unix shell and access all my dropbox files in mostly the same way as I do from my desktop machines. And I very much enjoy having all that, together with all the social media apps, calls, SMS, and a 4G connectivity. There are, of course, limitations. The Gemini is not the suitable device for writing your thesis, or your next 5,000 lines of code long program. It is possibly a device suitable for editing them, but not for creating them. 

So, do I consider the Gemini a mini-laptop ? "Well, I'm not sure" is probably an honest answer. Sometimes it does feel like a laptop, but keep in mind that I'm an old guy that works exclusively with GNU/linux, and then again, mostly from the terminal. I am a typist, and the physical keyboard is my natural working environment. On the other hand, when switching to the pure graphical android apps, the device feels mostly like a funny smartphone. I say funny because with apps like twitter, facebook, instagram, you have to turn the device in portrait mode and practically give-up the keyboard controls.  A natural question at this point is why haven't you install a proper linux distribution on the device and be done with it ? Two reasons. Actually three. The first was that I wanted the device to remain portable, I didn't want to have to carry with me a mouse. The second was that I would have to guess-estimate the needed partition sizes (and get them wrong). But the important reason is that I am conservative: I wanted to first test it in a pure android environment and then if needed become adventurous. After almost two months I believe that I can not see a real reason for messing with a dual boot setup. I would love to be able to do everything from the keyboard, but even in a linux environment I would have to either touch the screen or carry additional hardware with me.

As you almost certainly noticed, I haven't said anything about using the Gemini as my one and only phone. The reason is that I may misguide you on that one. I hardly speak on the phone for more than a 30 minutes per week, and I mostly don't really care about carefully examining who is calling (which means that normally I don't even open the device to accept the call). I understand that this is not the way things go these days, so I will just state for the record that my phone-related needs are served adequately by the device and leave it there.

To conclude. The Gemini is not -and was not meant to be- for the average, always in a hurry, smartphone user. It will not kill iphones or the other android devices. As another reviewer aptly said, it is a device for those that saw it and their heart skipped a beat. To paraphrase Thomas Scoville, the Gemini is for those users that love (instead of resent) typing. And for those users, the Gemini (like UNIX) will, in some sense, be literature to them.

Gemini PDA

I'm hooked.

As a reviewer was saying, if you skipped a beat when you first saw it, then you belong to the target audience of the Gemini PDA :

The bloody thing can even dual boot into Debian. Expect a review to appear here soon (hopefully).

Primitive cluster monitoring for the desktop

Combine warewulf, a tiny cgi perl script and conky, and there it is, a primitive cluster monitoring tool for the desktop without ssh :

Renewing cluster hardware

After almost nine years, the hardware from the old norma cluster (circa 2009) is useless for all practical purposes. Have been thinking about the possibility of preserving its (still) functional part and only upgrade the GPUs. Given that the CPUs are weak (Q6600 Kentsfield 2.4 GHz quad processors) I thought I could try with a cheap GPU and see how it goes. Adding a GTX1070 wouldn't make sense, so the first attempt was a GTX1050. The golden standard for comparing the performance is scarlet (i7-6800K@3.4GHz + GTX-1070). The findings were not surprising (system with 9418 atoms and a 4 fs/step using HMR) :

• Scarlet (i7-6800K @ 3.4GHz + GTX-1070)                 ==>  445 ns/day
• Old AMD FX-8150, 8 cores @ 3.6 GHz + GTX 1050  ==>  195 ns/day
• Old Q6600 @ 2.4GHz box + GTX 1050                      ==>  135 ns/day

Given the current price of a GTX1050 (~190 €) would it may make sense to order a handful of 1050s ? An additional test is to replace the boxes altogether with 8 core AMDs. Have ordered a relatively inexpensive box containing an AMD FX 8350 @ 4 GHz + a GTX 1050. The results were consistent with cost : at approximately a third of the price, you get half the performance (225 ns/day for a 9500 atom system on the new box, versus 445 ns/day on the i7+GTX-1070 box).

New fat compute node arrived

New fat compute node arrived. It contains an i7-6800K@3.4GHz, an nvidia GTX-1070, 16 GBytes of physical memory and two WD Red 1TB disks. After a few problems with setting-up the disks as RAID0 (using Intel's rapid storage), a minimal CentOS 7.3 was installed and the nvidia drivers compiled and installed.

First tests with NAMD 2.12 were impressive : on a small peptide problem with 4400 atoms, the new box delivered ~330 nanoseconds per day, a speed-up by a factor of 10 compared with NAMD 2.9 running on a node containing an 8-core AMD FX-8150 plus a GTX 570. To make the comparison hardware-dependent-only, we should test with NAMD 2.9 on the new box (testing new NAMD on the old box won't do because support for fermi GPUs was dropped with NAMD 2.12).

On the same peptide system (4300 atoms) and with a 2.5 fs time step, 470 ns/day were recorded. This reached an outstanding 550 ns/day for a smaller (3300 atoms) system. Think about it again : more than 1 μs in two days. Two days.

Update : It seems that with this hardware and NAMD 2.12 it is necessary to disable off-loading FFT to the GPU by using the 'usePMECUDA no' configuration option.

New server arrived

Our new IBM X3755 M3 server arrived. With 4 x 12-core AMD processors, it has as many cores in a single 2U box as the whole of our Beowulf cluster. Trouble is I do not know where to start with this beast (not to mention where and how to place it in the first place).

Using HP proliant microservers as students' terminals

I still can't believe it : 219€ (including 23% VAT) for a Hewlett Packard's HP Proliant Microserver:

Got two of them to be used as student terminals, setup one of them with linux Mint, use dd to clone the disk for the other, end of story.

ps. And it is getting even better : new price is 185,36 € (incl. VAT). Incredible.

New machine for cluster

New box with an 8-core AMD FX-8150, an NVIDIA GTX 570 and 4 Gbytes of memory at 1333 MHz arrived. After a few problems with perceus (rejected hardware address starting from 8C:…), all else appear to have worked smoothly. Started testing it with NAMD ... [results]

Will we get this ?

• 8 x Six-Core AMD Opteron™ Model 8435 - 2.6GHz 6MB L3 Cache
• 16 x 2GB PC2-6400 800MHz DDR2 ECC Registered
• 2 x 2.0TB SATA 6.0Gbps 7200RPM - 3.5" - Seagate Constellation™ ES
• PNY NVIDIA Quadro 2000 1.0GB GDDR5 (1xDVI-DL, 2xDP)

Full details ...

New GPU : Gigabyte's GTX 460

New nvidia GTX 460 installed on cluster. This is the gigabyte version of the card with a double fan. Very nice :