DCS4

Hard ware

This PC is ordered from OriginPC.

Item	Description
CPU	AMD Ryzen 9 9950X3D 16-Core Processor (up to 5.7 GHz)
Motherboard	MSI PRO B850-P WIFI7 (WIFI antanna not installed)
RAM	2 x 48 GB DDR5 6000MT/s
Cooling	Liquid cool
GPU	NVIDIA RTX PRO 2000 Blackwell 16 GB
Disk	1TB M.2 (OS) + 2x 8TB M.2 (storage)
network card	Intel X550-T2 10Gbps (2 ports)
Fiber Optics	CAEN A3818 Fiber optics adaptor

CAEN A3818 PCIe CONET2 Controller — Setup on dcs4

Reconstructed from the shell history on dcs4.onenet and verified against the live system on 2026-09-09. Every command below was actually run on this host.

Item	Value
Host	dcs4.onenet
OS	Ubuntu 24.04.4 LTS (noble)
Running kernel	7.0.0-30-generic
Card	CAEN A3818 (PCIe, 4× CONET2 optical links)
PCI address	05:00.0
PCI ID	10ee:1015 (Xilinx vendor, CAEN device); subsystem 10ee:0007
Driver	a3818 v1.6.12, installed via DKMS
DKMS source	/usr/src/a3818-1.6.12
Driver tarball	~/Desktop/CAEN_Digitizer_Libs/a3818_linux_driver-v1.6.12.tar.gz
Manuals	~/Desktop/CAEN_Manual/A3818/
Device nodes	/dev/a3818_<N> (see #Device node numbering)

Current state: the DKMS module is built for kernel 7.0.0-31-generic, but the machine is running 7.0.0-30-generic. The module is therefore not loaded and /dev/a3818_* does not exist. See #Module not loaded after a kernel update.

1. Prerequisites

The A3818 driver is an out-of-tree kernel module built by DKMS, so the toolchain and the headers for the running kernel must be present:

sudo apt update
sudo apt install dkms build-essential linux-headers-$(uname -r)

This was the cause of the first failed install attempt in the history — install.sh was run before dkms and the kernel headers were present, and the dkms build step aborted. Install these first.

2. Confirm the card is seen by the bus

Before touching the driver, check that the PCIe card enumerates. The A3818 presents itself as a Xilinx memory controller, not as anything named "CAEN":

lspci -nn | grep 10ee

Expected output on this host:

05:00.0 Memory controller [0580]: Xilinx Corporation Device [10ee:1015]

If this line is absent, stop — the problem is hardware/seating/BIOS, not software. No amount of driver work will help.

3. Unpack the driver

The driver ships inside the CAEN library bundle:

cd ~/Desktop/CAEN_Digitizer_Libs
tar -xvf a3818_linux_driver-v1.6.12.tar.gz
cd a3818_linux_driver-v1.6.12

Contents:

File	Purpose
src/a3818.c, src/a3818.h	Driver source
install.sh	DKMS install wrapper (the supported path)
Makefile	Manual build (make / sudo make install)
dkms.conf.in	DKMS template; install.sh substitutes the version into dkms.conf
dkms_post_add.sh	Writes /etc/udev/rules.d/10-CAEN-A3818.rules
dkms_post_remove.sh	Deletes that udev rule again
Readme.txt, ReleaseNotes.txt	CAEN docs

If a previously-extracted a3818_linux_driver-v1.6.12/ directory is already present, delete it before extracting (rm -rf a3818_linux_driver-v1.6.12). A stale tree with a half-written dkms.conf will make install.sh fail in confusing ways.

4. Install the driver

chmod +x install.sh
sudo ./install.sh

Must be run as root — the script writes to /usr/src and calls dkms.

What install.sh does, in order:

1. Generates dkms.conf from dkms.conf.in with VERSION=1.6.12.
2. rmmod a3818 (ignored if not loaded).
3. dkms uninstall / dkms remove a3818/1.6.12 --all to clear any prior install.
4. Copies the tree to /usr/src/a3818-1.6.12.
5. dkms add → dkms build → dkms install --force.
6. modprobe a3818.

Because dkms.conf.in sets AUTOINSTALL=yes, DKMS will rebuild the module automatically for each new kernel installed by apt from this point on.

Manual (non-DKMS) alternative

Only if DKMS is unavailable. Note this build will not survive a kernel update:

make
sudo make install
# and to reverse it:
sudo make uninstall

5. Verify the module loaded

lsmod | grep a3818
dkms status

dkms status should report the module as installed for the running kernel:

a3818/1.6.12, 7.0.0-31-generic, x86_64: installed

Kernel messages from the probe:

sudo dmesg | grep -i a3818
# CAEN PCIe: N device(s) found.

dkms status prints Deprecated feature: REMAKE_INITRD. This comes from CAEN's dkms.conf and is harmless — the A3818 module is not needed in the initrd.

The driver also registers a proc entry:

cat /proc/a3818

6. udev rule for non-root access

By default the character devices are root-owned, so CoMPASS and any user-space program using CAENComm would need sudo. The driver registers a device class named a3818 and creates nodes a3818_<N> under it, so a udev rule can widen the permissions.

The rule in use on this host is /etc/udev/rules.d/99-caen-a3818.rules:

sudo nano /etc/udev/rules.d/99-caen-a3818.rules

# CAEN A3818 PCIe CONET2 controller - open access to all local users
SUBSYSTEM=="a3818", KERNEL=="a3818_*", MODE="0666"

Reload and re-trigger:

sudo udevadm control --reload-rules
sudo udevadm trigger --subsystem-match=a3818
ls -l /dev/a3818_0

Relationship to CAEN's own rule

dkms_post_add.sh writes its own rule at /etc/udev/rules.d/10-CAEN-A3818.rules:

KERNEL=="a3818_[0-9]", MODE="0666"
KERNEL=="a3818_[0-9][0-9]", MODE="0666"

and dkms_post_remove.sh deletes it again. Because install.sh runs dkms remove before dkms add, a re-install cycle can leave this file missing — and indeed it is not present on dcs4 right now; only the hand-written 99- rule exists. The 99- rule is the more robust of the two (it matches any index, not just one or two digits), so keeping it is deliberate. If you ever reinstall the driver and find both files, that is fine — they set the same mode.

MODE="0666" grants read/write on the card to every local user. That is acceptable on a dedicated DAQ box like dcs4. On a shared machine, prefer GROUP="caen", MODE="0660" and add DAQ users to a caen group.

7. Device node numbering

The driver computes the minor number, and hence the node name, as:

index = (CardNumber << 6) + (OpticalLink << 3) + V2718Slave
/dev/a3818_<index>

Node	Card	Link	Slave
/dev/a3818_0	0	0	0
/dev/a3818_8	0	1	0
/dev/a3818_16	0	2	0
/dev/a3818_24	0	3	0
/dev/a3818_64	1	0	0

So a single A3818 with its four CONET2 links produces nodes at 0, 8, 16 and 24 (plus the daisy-chain slave offsets in between). When a CAENComm/CAENDigitizer program asks for "link 0", it opens /dev/a3818_0.

8. CAEN user-space libraries

The kernel driver alone is not enough — CoMPASS and any DAQ code need the CAEN shared libraries. These are installed by the bundle's own script:

cd ~/Desktop/CAEN_Digitizer_Libs
chmod +x Install_libraries.sh
sudo ./Install_libraries.sh

It installs build dependencies (build-essential, libusb-1.0-0-dev, qt6-base-dev, libqt6charts6-dev, libcurl4-openssl-dev, libboost-all-dev, libgl1-mesa-dev) and then builds, in order:

Library	Version on dcs4	Installed to
CAENVMELib	v4.1.3	/usr/lib/libCAENVME.so
CAENComm	v1.8.0	/usr/lib/libCAENComm.so
CAENDigitizer	v2.19.0	/usr/lib/libCAENDigitizer.so
CAEN FELib	v1.3.2	/usr/local/lib/libCAEN_FELib.so
CAEN Dig2	v1.8.3	/usr/local/lib/libCAEN_Dig2.so
CAEN Dig1 (binary)	v1.1.4	/usr/local/lib/libCAEN_Dig1.so

Verify:

ldconfig -p | grep -i caen

Useful flags at the top of Install_libraries.sh:

• INSTALL_A4818=false — set true only if the USB A4818 bridge is also in use. Not needed for the A3818.
• BOOST_PATH="" — point at a manually-extracted Boost if the system Boost is too new for caen_dig2.
• sudo ./Install_libraries.sh clean — removes the extracted source directories but keeps the tarballs.

The A3818 kernel driver is deliberately not handled by Install_libraries.sh. It is installed separately, as in #4. Install the driver.

9. Troubleshooting

Module not loaded after a kernel update

This is the current state of dcs4. Symptoms: lsmod | grep a3818 is empty, modinfo a3818 says "Module a3818 not found", /dev/a3818_* does not exist — yet lspci still shows the card.

Cause: DKMS built the module for the newest installed kernel, not the running one:

dkms status
# a3818/1.6.12, 7.0.0-31-generic, x86_64: installed
uname -r
# 7.0.0-30-generic

Two fixes:

# Preferred: reboot into the kernel the module was built for
sudo reboot

# Or build for the currently running kernel (headers for it are installed)
sudo dkms install a3818/1.6.12 -k $(uname -r) --force
sudo modprobe a3818

Confirm afterwards with lsmod | grep a3818 and ls -l /dev/a3818_0.

"Operation not allowed" when inserting the module

Secure Boot is enabled and the kernel is rejecting the unsigned out-of-tree module. Either sign the module with an enrolled MOK, or disable Secure Boot in the BIOS. See the N.B. in CAEN's Readme.txt.

dkms build fails

Almost always missing headers for the running kernel:

sudo apt install dkms build-essential linux-headers-$(uname -r)

Then re-run sudo ./install.sh. The build log is at /var/lib/dkms/a3818/1.6.12/build/make.log.

No /dev/a3818_* even though the module is loaded

The driver only registers the char device if the PCI hardware is found — this is deliberate, to avoid a kernel crash on cat /proc/devices. Check lspci -nn | grep 10ee and sudo dmesg | grep -i a3818 for the "CAEN PCIe: N device(s) found." line.

Permission denied opening the card from a user program

The udev rule is missing or was not applied. Re-run:

sudo udevadm control --reload-rules
sudo udevadm trigger --subsystem-match=a3818
ls -l /dev/a3818_0    # should be crw-rw-rw-

10. Quick reference

# Is the card on the bus?
lspci -nn | grep 10ee

# Is the driver loaded and built for this kernel?
lsmod | grep a3818
dkms status
uname -r

# Device nodes and permissions
ls -l /dev/a3818_*
cat /proc/a3818

# Reload the module
sudo rmmod a3818 && sudo modprobe a3818

# Reapply udev rules
sudo udevadm control --reload-rules
sudo udevadm trigger --subsystem-match=a3818

# Rebuild for the running kernel
sudo dkms install a3818/1.6.12 -k $(uname -r) --force

# Full reinstall from source
cd ~/Desktop/CAEN_Digitizer_Libs/a3818_linux_driver-v1.6.12 && sudo ./install.sh

# Remove entirely
sudo dkms uninstall a3818/1.6.12
sudo dkms remove a3818/1.6.12 --all

# CAEN user-space libraries
ldconfig -p | grep -i caen

See also

• ~/Desktop/CAEN_Manual/A3818/ — CAEN A3818 hardware manual
• ~/Desktop/CAEN_Digitizer_Libs/UM1934_CAENComm_User_Manual_rev.9.pdf
• ~/Desktop/CAEN_Digitizer_Libs/UM1935_CAENDigitizer Library_rev.22.pdf
• ~/Desktop/CAEN_Digitizer_Libs/UM7715_CAENVMELib_User_Manual_rev.2.pdf

Software

Inventory and setup notes for the analysis / DAQ / monitoring software on dcs4.onenet. Reconstructed from the shell history and verified against the live system on 2026-09-09.

Four of the seven packages below are not yet installed. The status column is the short answer; each section gives the install procedure either way. See #Installation status.

Installation status

Software	Purpose	Version
CERN ROOT	Analysis framework	6.36.06
InfluxDB 3 Core	Time-series database	3.11.2
Grafana	Dashboards	13.2.1 Enterprise
CoMPASS	CAEN turnkey DAQ	v2.6.4 available
FSUDAQ	DAQ for 1st-gen CAEN digitizers	—
SOLARIS DAQ	DAQ for 2nd-gen CAEN digitizers	—
ELOG client	Electronic logbook	3.1.5 upstream

The CAEN A3818 driver and user-space libraries that the DAQ programs depend on are documented separately — see the A3818 page.

Dependency map

                      A3818 kernel driver (v1.6.12, DKMS)
                                  |
        +-------------------------+-------------------------+
        |                         |                         |
   CAENVMELib 4.1.3          CAEN FELib 1.3.2          (USB / A4818)
   CAENComm   1.8.0          CAEN Dig2  1.8.3
   CAENDigitizer 2.19.0      CAEN Dig1  1.1.4
        |                         |
     FSUDAQ                  SOLARIS DAQ  ------> ROOT (EventBuilder only)
     CoMPASS  (bundles its own ROOT + Qt + JDK)
        |                         |
        +---------> InfluxDB 3 <--+ ---------> Grafana
        +---------> ELOG <--------+

Both FSUDAQ and SOLARIS DAQ can push scaler rates to InfluxDB and post entries to ELOG, which is why those four pieces belong on one page.

CERN ROOT

Item	Value
Version	6.36.06
Install path	/opt/root_v6.36.06, with /opt/root as a symlink
Distribution	Official pre-built binary, ubuntu24.04-x86_64-gcc13.3
Activated by	source /opt/root/bin/thisroot.sh in ~/.bashrc (line 159)

Build dependencies

ROOT is installed from the pre-built binary, so these are not strictly required to run it — but they are needed to compile anything against ROOT, and were installed here:

sudo apt install binutils cmake dpkg-dev g++ gcc libssl-dev git libx11-dev libxext-dev \
  libxft-dev libxpm-dev python3 libtbb-dev libvdt-dev libgif-dev

sudo apt install gfortran libpcre3-dev libglu1-mesa-dev libglew-dev libftgl-dev libfftw3-dev \
  libcfitsio-dev libgraphviz-dev libavahi-compat-libdnssd-dev libldap2-dev python3-dev \
  python3-numpy libxml2-dev libkrb5-dev libgsl-dev qtwebengine5-dev nlohmann-json3-dev \
  libmysqlclient-dev libgl2ps-dev liblzma-dev libxxhash-dev liblz4-dev libzstd-dev \
  libcurl4-openssl-dev

Install

cd ~/Downloads
tar -xvf root_v6.36.06.Linux-ubuntu24.04-x86_64-gcc13.3.tar.gz
mv root root_v6.36.06
sudo mv root_v6.36.06 /opt/.
cd /opt
sudo ln -s root_v6.36.06 root

The versioned directory plus a root symlink is deliberate: upgrading means unpacking the new version alongside and re-pointing the symlink, with no change to ~/.bashrc and an instant rollback if the new build misbehaves.

Environment

In ~/.bashrc:

#================================== CERN ROOT
source /opt/root/bin/thisroot.sh

This sets ROOTSYS, PATH, LD_LIBRARY_PATH and PYTHONPATH. Reload with rebash (alias for source ~/.bashrc).

Verify

root-config --version      # 6.36.06
root-config --features
root -l -q                 # starts and exits cleanly

Features compiled into this build include pyroot, dataframe, roofit, tmva (with tmva-cpu, tmva-sofie), fftw3, fitsio, spectrum, unfold, gdml, opengl, http, sqlite, fortran and cxx17. Cling and LLVM are the builtin versions.

ELOG client

PSI's Electronic Logbook. Both FSUDAQ and SOLARIS DAQ can post run entries to an ELOG server, using the elog command-line client. Current upstream version is 3.1.5.

Not installed, and not installable via apt on this machine. FSUDAQ's README says sudo apt install elog, which was correct on Ubuntu 22.04. The elog package does not exist in Ubuntu 24.04 (noble) — apt-cache policy elog returns nothing here even though universe is enabled. It must be built from source.

Build from source

sudo apt install cmake build-essential git
git clone https://gitea.psi.ch/ELOG/elog --recursive
cd elog
mkdir build && cd build
cmake ..
make

This produces two binaries in the build directory:

Binary	Role
elog	Command-line client — this is the one FSUDAQ/SOLARIS need
elogd	The logbook server daemon

There is no make install target; place the binaries manually:

sudo cp elog /usr/local/bin/
# only if this host is also to run the logbook server:
sudo cp elogd /usr/local/sbin/

Upstream uses no release tags — build from master.

dcs4 only needs the client if the logbook lives on another machine. Decide that before setting up elogd, a config file and a systemd unit.

Client usage

elog -h <host> -p <port> -l <logbook> -u <user> <password> \
     -a Author="Ryan" -a Type="Run" -a Run=123 \
     "Run 123 started, 4 digitizers armed"

The same host/port/logbook/user/password values go into lines 5–8 of FSUDAQ's programSettings.txt.

Docs: https://elog.psi.ch/elog/ · PDF manual: https://elog.psi.ch/elog/pdf/document.pdf

InfluxDB 3 Core

Time-series database, used as the sink for scaler rates and slow-control data from the DAQ, and as Grafana's data source.

There is already a full operations document for this at ~/influx.md — tokens, databases, writing, querying, deletion limitations, troubleshooting. This section is only the summary. Read that file for anything beyond the basics: mdless ~/influx.md.

Item	Value
Product	InfluxDB 3 Core 3.11.2
Binary	~/.influxdb/influxdb3
Data dir	~/.influxdb (object store: local file)
Node ID	dcs4.onenet-node
HTTP API	http://0.0.0.0:8181
Service	influxdb.service — enabled, active
Runs as	adminrt
Admin token	~/influx.token (mode 600)
Databases	MUSIC, _internal

Install

curl -O https://www.influxdata.com/d/install_influxdb3.sh && sh install_influxdb3.sh
source ~/.bashrc
rm install_influxdb3.sh
influxdb3 create token --admin        # save the output to ~/influx.token

Environment

In ~/.bashrc:

#================================== influxDB 3
export PATH="$PATH:/home/phy/adminrt/.influxdb/"
export INFLUXDB3_AUTH_TOKEN=$(sed -n 's/^Token: //p' ~/influx.token)

With INFLUXDB3_AUTH_TOKEN exported, the CLI needs no --token flag.

Service

/etc/systemd/system/influxdb.service:

[Unit]
Description=InfluxDB 3 Core Database Server
Documentation=https://docs.influxdata.com/influxdb3/core/
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=adminrt
Group=adminrt
LimitNOFILE=65536
EnvironmentFile=-/etc/default/influxdb3
ExecStart=/home/phy/adminrt/.influxdb/influxdb3 serve \
      --node-id dcs4.onenet-node \
      --object-store file \
      --data-dir /home/phy/adminrt/.influxdb \
      --http-bind 0.0.0.0:8181
Restart=on-failure
RestartSec=5s

[Install]
WantedBy=multi-user.target

Quick commands

sudo systemctl status influxdb.service
curl -i http://localhost:8181/health          # 401 without a token — that means it IS running
influxdb3 show databases
influxdb3 create database MUSIC
influxdb3 write -d MUSIC "testing value=1"

An unauthenticated GET /health returns 401 Unauthorized, not 200. That is the expected healthy response — it proves the server is up and enforcing auth. Pass -H "Authorization: Bearer $INFLUXDB3_AUTH_TOKEN" for a 200.

This is InfluxDB 3, not 1.x or 2.x. The daemon is influxdb3 serve, not influxd, and 1.x/2.x guides do not apply. This matters for FSUDAQ, which was written against 1.8 — see the warning in #FSUDAQ.

Grafana

Dashboards over the InfluxDB data.

Item	Value
Package	grafana-enterprise 13.2.1 (.deb)
Binary	/usr/sbin/grafana-server
Config	/etc/grafana/grafana.ini
Data	/var/lib/grafana
Share	/usr/share/grafana
Service	grafana-server.service — enabled, active
Web UI	http://dcs4.onenet:3000

Install

sudo apt-get install -y adduser libfontconfig1 musl
cd ~/Downloads
wget https://dl.grafana.com/grafana-enterprise/release/13.2.1/grafana-enterprise_13.2.1_33191028959_linux_amd64.deb
sudo dpkg -i grafana-enterprise_13.2.1_33191028959_linux_amd64.deb
sudo systemctl daemon-reload
sudo systemctl enable --now grafana-server.service
sudo systemctl status grafana-server.service

The grafana-enterprise package is the same binary as OSS Grafana; the enterprise features simply stay locked without a licence. No licence is installed here, so this is functionally the free build.

Configuration

/etc/grafana/grafana.ini is entirely at defaults — every line is commented out. That means port 3000, no TLS, and the stock admin/admin credentials with a change prompt at first login.

Change the admin password at first login. With the config untouched, Grafana is listening on 0.0.0.0:3000 with default credentials.

Verify

sudo systemctl status grafana-server.service
curl -s http://localhost:3000/api/health
# {"database":"ok","version":"13.2.1", ...}

Connecting to InfluxDB 3

Not yet provisioned — /etc/grafana/provisioning/datasources/ contains only the stock sample.yaml. To add the data source through the UI:

• Type: InfluxDB
• Query language: SQL (InfluxDB 3 native; not InfluxQL or Flux)
• URL: http://localhost:8181
• Database: MUSIC
• Token: Ask Ryan.

No need to use HTTP headers

CoMPASS

Source: CAEN SpA. Installed at ~/CoMPASS-v2.6.4, extracted from the tarball kept in ~/Desktop/CAEN_Digitizer_Libs/.

CAEN's multiparametric acquisition program. It is a sealed appliance: unpack, run, and take data — at the cost of being unable to change anything about how the data is processed.

Version	Bundled JDK	Plotter	Install
v2.6.4, January 2026	jdk-21.0.2	Qt 5.15.2 + ROOT 6.28.06	Extract in place, no compile

How CoMPASS was installed here

Extract and move. Nothing else.

cd ~/Desktop/CAEN_Digitizer_Libs
tar -xvf CoMPASS-v2.6.4.tar.gz
mv CoMPASS-v2.6.4 ~/
cd ~/CoMPASS-v2.6.4
./CoMPASS

install.sh was deliberately not run on this machine. It only does two things, both optional:

• symlinks the launcher to /usr/local/bin/CoMPASS so it is on $PATH — we run ./CoMPASS from the folder instead;
• copies a4818.rules to /etc/udev/rules.d and reloads udev, so an A4818 optical-USB adaptor is reachable without root — irrelevant while the digitizers are on Ethernet.

Run sudo ./install.sh only if you want one of those. Uninstalling is deleting the folder either way; there is no package manager state to unwind.

The older tarballs (CoMPASS-v2.1.1, v2.2.1) are still in CAEN_Digitizer_Libs/. Versions are fully parallel — extract one elsewhere and run it directly if you need to compare behaviour.

The g++-11 problem

This is the one thing that will stop a fresh CoMPASS on Ubuntu 24.04, and the launcher's own version check does not catch it.

Symptom. CoMPASS starts, the main window appears, but the plot window never does. ~/Compass/logs/log.0.txt reports Process CoMPASSPlot has been terminated with exit value 139 (SIGSEGV) followed by a Java IllegalStateException. The DPP server is fine — CAENDPPServerLog.txt shows it binding its ports and accepting the client, then logging a clean Client disconnection when the plotter takes the app down. CoMPASSPlotLog.txt contains only its logger-init line; the plotter dies before writing anything else.

Root cause. The ROOT/cling build bundled in plot/linux/ was compiled on Ubuntu 22.04 against GCC 11, and it hardcodes the compiler name x86_64-linux-gnu-g++-11 to discover the C++ standard-library include paths at runtime:

LC_ALL=C x86_64-linux-gnu-g++-11 -xc++ -E -v /dev/null 2>&1 | sed -n ...

On Ubuntu 24.04 that binary does not exist — only g++ 13.3.0 and /usr/include/c++/13. The pipeline still exits 0, because sed succeeds even though the compiler was not found, so cling never bails out cleanly. It proceeds with an empty include list, builds a VFS overlay entry with 'name': , and segfaults in llvm::vfs::OverlayFileSystem::pushOverlay(), inside plot/linux/lib/libCling.so.

The stack trace in log.0.txt shows the whole chain:

main
 -> RootApplication::RootApplication()
 -> TApplication::TApplication()
 -> TROOT::InitInterpreter()
 -> CreateInterpreter()                              [libCling.so]
 -> cling::Interpreter::Interpreter()
 -> cling::IncrementalParser::IncrementalParser()
 -> cling::CIFactory::createCI()
 -> setupCxxModules(clang::CompilerInstance&)
 -> collectModuleMaps(...)
 -> llvm::vfs::OverlayFileSystem::pushOverlay(...)    <-- SEGFAULT

Fix.

sudo apt install g++-11

This provides /usr/bin/x86_64-linux-gnu-g++-11 and /usr/include/c++/11, exactly what cling probes for. It installs alongside GCC 13 and does not change the system default compiler.

Why the launcher does not catch this. The CoMPASS script checks g++ -dumpfullversion against a minimum of 11.4.0. GCC 13.3.0 passes that check comfortably, so the launcher proceeds — while the thing cling actually needs, a binary literally named g++-11, is missing. The version check and the runtime dependency are testing different things.

If g++-11 alone does not fix it

ROOTSYS=/opt/root is set in the environment here, and /opt/root/lib gets appended to the LD_LIBRARY_PATH CoMPASS builds for itself. CoMPASS's own plot/linux/lib comes first so it should win, but a separate ROOT install can still feed the wrong etc/ modulemaps into collectModuleMaps. Next test:

env -u ROOTSYS ./CoMPASS

Launcher preflight checks

The CoMPASS script refuses to start rather than failing halfway, and each check prints its own reason. All of these currently pass on this machine.

Requirement	Minimum	How it is checked
Architecture	x86-64	uname -m must contain 64
Java	21	Bundled JDK, version parsed from -version
g++	11.4.0	g++ -dumpfullversion, compared field by field
glibc	2.34	strings on the libc that CAENDPPServer links
glibcxx	3.4.30	strings on the libstdc++ that CoMPASSPlot links

CoMPASS is three processes, not one

What looks like a single application is a Java front end talking over local TCP to a native back end, plus a separate plotting binary. Knowing this is what makes the logs readable — and it is why a plotter crash leaves a perfectly healthy digitizer log.

Process	Role	Endpoint
compass-v2.6.4.jar	Java 21 GUI, run configuration, orchestration	—
CAENDPPServer	Native digitizer control and readout	127.0.0.1:56342 control 127.0.0.1:56343 data
CoMPASSPlot	Qt5 + ROOT histogram and waveform display	—

Those ports are set by the launcher via -Ddpp.server.* flags, and the server is spawned by the JVM (dpp.server.launch=true). If a stale CAENDPPServer is holding 56342, a new CoMPASS will not come up — check ~/Compass/logs/CAENDPPServerPID and kill it.

Library collision. server/linux/ ships CAEN's whole stack — libCAENDigitizer, libCAEN_Dig2, libCAEN_FELib, libCAENComm, libCAENVME — and prepends that directory to LD_LIBRARY_PATH. It is deliberately insulated from the system CAEN libraries that FSUDAQ and SOLARIS_DAQ link against, so the two can disagree on version without either breaking. It also means a system library upgrade does nothing for CoMPASS.

CoMPASS files and logs

The working directory is ~/Compass/, created on first run, and separate from the install directory.

Path	Contents
~/Compass/config/	Saved project and board configurations
~/Compass/data/	Acquired runs
~/Compass/logs/log.N.txt	Java front-end log, rotated; log.0.txt is current — start here
~/Compass/logs/CAENDPPServerLog.txt	Native server — board and link errors
~/Compass/logs/CAENDPPServerPID	PID of the running native server
~/Compass/logs/CoMPASSPlotLog.txt	Plotter, including ROOT complaints

Documentation ships in the install folder rather than online: UM5960_Compass_UserManual_rev27.pdf is the reference manual and GD6300_Compass_QuickStart_rev27.pdf the getting-started guide. Older revisions (rev19–rev23) sit in ~/Desktop/CAEN_Manual/, filed per board. releasenotes.txt carries the full changelog.

FSUDAQ

Source: github.com/goluckyryan/FSUDAQ. Cloned to ~/FSUDAQ.

A Qt6 acquisition system for first-generation CAEN digitizers, with a half-second scope, full register control, and an online analysis layer you are expected to write against.

Boards	Firmware	Binary	Data format
x725, x725S, x730, V1740	PHA, PSD, QDC	FSUDAQ_Qt6	.fsu

Everything on the digitizer is reachable from the settings panel except LVDS. The scope redraws every 500 ms, waveforms can be saved, and each channel carries its own 1D histogram — off by default, switched on in the Online Histograms panel, with the binning persisted to singleSpectaSetting.txt under the raw data path.

FSUDAQ dependencies

The CAEN libraries must be downloaded and installed by hand; the rest come from apt.

Component	Minimum	Needed for
CAENVMELib	3.3 (4.0+ for A5818)	VME access layer
CAENComm	1.5.3	Link abstraction
CAENDigitizer	2.17.1	Board control
CAEN A3818 driver	1.6.8	PCIe optical card
CAENUSBdrv	1.5.5	V57xx over USB
Qt6 + Charts	—	GUI and plots
libcurl	—	InfluxDB writes
elog	3.1.3+	Logbook entries

sudo apt install qt6-base-dev libqt6charts6-dev libcurl4-openssl-dev elog

QCustomPlot is vendored in the repository, so it is not an external dependency.

Building FSUDAQ

cd ~/FSUDAQ
qmake6 FSUDAQ_Qt6.pro     # regenerates the Makefile
make -j$(nproc)
./FSUDAQ

Re-run qmake6 whenever the .pro changes — notably after adding an analyzer. For a debug build, uncomment QMAKE_CXXFLAGS += -g and the two -O0 lines in the .pro; the program installs an abort handler that writes a core dump, which needs ulimit -c unlimited and a writable core_pattern to land anywhere useful.

Optical link handling

A4818 cabling order. Connect the optical fibre before powering the digitizers on. If a fibre is unplugged and reconnected during a session, the digitizers must be power-cycled — reconnecting alone will not restore the link.

A Raspberry Pi 5 with 8 GB and an A4818 is a tested configuration; all the CAEN libraries build for ARM. On Pi, substitute qt6-chart-dev for libqt6charts6-dev, and install elog by hand or skip it.

FSUDAQ settings and data

programSettings.txt sits beside the binary and is read line by line — position is the only key, so blank lines are meaningful placeholders and must be kept. The copy in ~/FSUDAQ is currently blank throughout, so no data path, InfluxDB or elog is configured.

#	Field	Effect if empty
1	raw data path	Can still run ACQ, but cannot save a run
2	InfluxDB IP	InfluxDB logging disabled
3	database name
4	InfluxDB token (v2.0+)
5	elog IP	elog disabled
6	elog logbook
7	elog user
8	elog password

With no settings file at all, FSUDAQ still starts and still discovers every digitizer on the optical link. Degradation is graceful and per-feature.

What lands in the data path

File	Purpose
*.fsu	Raw acquired data
*.bin	Digitizer settings snapshot for the run
RunTimestamp.dat	Start/stop times and comments for every run
RunTimestamp.csv	The same, for a spreadsheet
lastRun.sh	Last-run values exported for shell scripts
singleSpectaSetting.txt	Per-channel histogram binning

Compiled-in limits

Limit	Value
Digitizers	16 (4 ports × 4 boards)
Channels	64
Record length	0x3FFF × 8
Max file size	2 GB
Buffer per channel	10,000 events
Traces per event	5
Scope refresh	200 ms
Scaler refresh	1,000 ms

Keyboard control

The panels are fully keyboard-drivable, which matters when the DAQ box is on a KVM at the far end of a hall.

Key	Action
F4	Open the focused drop-down
Tab / Ctrl+Tab	Next / previous element
Space	Press the focused button
↑ / ↓	Increment / decrement a number
Alt+Tab	Switch between DAQ windows

Field colour is state, not decoration: blue means the value has been typed but not yet pushed to the board; red means the board rejected it.

FSUDAQ online analyzers

Read this before debugging empty spectra. No events are built when FSUDAQ starts. The MultiBuilder only begins once an online analyzer has been created and opened — and once it has, event building continues even if you close the analyzer window. An empty coincidence spectrum usually means nobody opened an analyzer, not that the physics is wrong.

MultiBuilder correlates hits across several digitizers. It builds forward in the usual way, and also backward — from the most recent data back through a fixed number of events — which is what lets the online plots stay responsive at high rate instead of lagging behind the front of the buffer.

Above that sit 1D and 2D histograms with graphical cuts that can be drawn, named, and renamed, and whose rates can be pushed to InfluxDB.

Analyzers in the repository

Header	Purpose
Analyser.h	Base class — subclass this
SplitPoleAnalyzer.h	Reference example: a 2D and a 1D histogram, plus cut rates to InfluxDB
CoincidentAnalyzer.h	Coincidence conditions
MUSICAnalyzer.h	MUSIC detector
NeutronGamma.h	n/γ pulse-shape discrimination
EncoreAnalyzer.h	Encore array
RAISOR.h	RAISOR separator
Isotope.h, Cross.h	Mass table (mass20.txt) and kinematics helpers

Adding your own analyzer

Derive from Analyser.h, then wire it into four places — the analyzer will not appear in the menu until all four are done:

1. Add the source file to FSUDAQ_Qt6.pro.
2. Include the header at the top of FSUDAQ.cpp.
3. Add an entry to the onlineAnalyzerList vector, also at the top of FSUDAQ.cpp.
4. Handle it in FSUDAQ::OpenAnalyzer().

qmake6 FSUDAQ_Qt6.pro
make

FSUDAQ known issues

These are documented upstream and are worth reading as operating procedure rather than as a bug list.

Aggregation settings will hang the board

Correlated registers. Aggregate Organization, Events per Aggregate and Record Length constrain one another. Invalid combinations are accepted by the panel and then send the digitizer into an unrecoverable state. When using the scope, Agg/Read must stay below 512.

For PHA firmware, Events-per-Aggregate is recomputed through the CAEN API before ACQ starts, but that calculation assumes the current rate — if the trigger rate changes substantially mid-experiment, it needs setting again by hand. Trapezoid scaling and the fine-gain register are likewise computed at ACQ start.

Everything else

Symptom	Cause / action
Negative trigger or accept rate	Buffer arriving out of time order. A digitizer settings problem, not a program bug — reprogram the board.
Pile-up rate looks wrong	Inaccurate above roughly 60 kHz input.
GUI stalls	Database access is slow. Disable InfluxDB.
Loaded settings look incomplete	Expected. Loading restores channel settings and some board settings only.
V1740D QDC: ReadData error, ACQ stops	When the first channel group is enabled, channel 0 must also be enabled.
V1740 QDC record length behaves oddly	Record length is a board-level setting, but readout is per group.
EventBuilder crashes offline	Known: very high trigger rates in the data.

SOLARIS_DAQ

Source: github.com/goluckyryan/SOLARIS_DAQ. Cloned to ~/SOLARIS_DAQ.

Built for the SOLARIS solenoidal spectrometer at FRIB and the x2730-series digitizers. Second-generation boards speak Ethernet and a REST-like endpoint model rather than VME registers, and the program is structured around that difference.

Boards	Firmware	Max digitizers	Data format
VX2730, VX2740, VX2745	DPP-PHA, DPP-PSD	20	.sol

The digitizer control classes are deliberately independent of the Qt layer, so the same core can be driven from the auxiliary command-line tools without dragging in a GUI.

Core, independent of Qt

File	Role
ClassDigitizer2Gen.h/cpp	Connection, configuration, readout, file I/O
Hit.h	Decoded event structure
RawDecoder.h	Unpacks the raw endpoint blob into hits
DigiParameters.h	Register definitions for both firmwares
RingBuffer.h	Lock-free circular buffer for per-channel spectra

Qt layer

File	Role
mainwindow.h/cpp	Digitizer management, run control, scalers
digiSettingsPanel.h/cpp	Board and channel register editor
scope.h/cpp	Waveform display
SingleSpectra.h/cpp	Per-channel energy spectra
SOLARISpanel.h/cpp	SOLARIS detector mapping and configuration
CustomThreads.h	ReadDataThread and TimingThread
ClassElog.cpp, ClassInfluxDB.cpp	Logbook entries and scaler-rate logging

Building SOLARIS_DAQ

cd ~/SOLARIS_DAQ
qmake6 SOLARIS_DAQ.pro
make -j$(nproc)
./SOLARIS_DAQ

Missing header. CAENDig2.h is not installed to the system include path by the CAEN Dig2 package. Copy it out of the source tree before the first build:

sudo cp caen_dig2-vXXXX/include/CAENDig2.h /usr/local/include/

Tested against V2740 and V2745 DPP-PHA-1G and DPP-PSD-1G, and VX2730 DPP-PSD from firmware 2025052203.

SOLARIS readout formats

Chosen through SetDataFormat(). Each step down the table drops metadata in exchange for bandwidth, and the choice is the single biggest lever on sustainable rate.

Format	ID	Carries
ALL	0x00	All metadata, 2 analog probes, 4 digital probes
OneTrace	0x01	1 analog probe + energy, timestamp, flags
NoTrace	0x02	Energy, timestamp, flags — no waveform
Minimum	0x03	Channel, energy, timestamp
MiniWithFineTime	0x04	As above plus fine timestamp
Raw	0x0A	Undecoded blob, unpacked in software

What a decoded hit holds

Field	Width	Units
channel	6 bit	—
energy	16 bit	ADC
energy_short	16 bit	ADC — PSD only
timestamp	48 bit	ns
fine_timestamp	10 bit	ps
flags, high / low priority	8 / 12 bit	—
analog probes ×2	18 bit/sample	—
digital probes ×4	1 bit/sample	—

What SOLARIS_DAQ pushes to InfluxDB

Six measurements, in two groups. Points accumulate in one buffer and are flushed as a single HTTP POST per tick, so a 64-channel board sends 128 points every 2 s in one request. No timestamps are sent on any point — the server stamps them at write time.

Continuous, one batch every 2 s

Measurement	Tags	Field	Source
Rate	Bd=<serial>, Ch=<NN>	value	PHA::CH::SelfTrgRate, read straight off the board
AccpRate	Bd=<serial>, Ch=<NN>	value	Computed in software — see below
FileSize	—	value	Sum of GetTotalFilesSize() across all digitizers, in bytes

Ch is zero-padded to two digits (mainwindow.cpp:1372). The 2 s cadence is TimingThread's default of 20 × 100 ms — the SetWaitTimeSec call at mainwindow.cpp:78 is commented out, so it is never overridden. FileSize is additionally gated on the Save Run checkbox (mainwindow.cpp:1380).

Event-driven

Measurement	Tags	Fields	Fires on
ProgramStart	—	value=1	InfluxDB setup succeeding — doubles as the write test
ProgramStart	—	value=0	Program exit
StartStop	—	value=1 or 0	Run start / stop; also scope open, and the scope's ACQ toggle
RunID	start=1	value=<runID>, expName, comment	Run start
RunID	start=0	value=<runID>, expName, comment	Run stop, new experiment, experiment switch

Spaces in comments are replaced with underscores before writing.

Two gates that silently stop the rate data

The Scaler window must be open. UpdateScalar() begins with if( scalar->isVisible() == false ) return; (mainwindow.cpp:1318), before any InfluxDB work happens. Close that window and Rate, AccpRate and FileSize stop dead — with no warning, and with RunID and StartStop still flowing normally because they sit on separate code paths. A Grafana panel can therefore go flat while the run looks perfectly healthy.

The second gate is scalarOutputInflux, set true at run start and false at run stop (mainwindow.cpp:525, 589), so the continuous group only flows during a run.

How AccpRate differs from Rate

Rate is the firmware's self-trigger rate register, and does not apply the coincidence condition.

AccpRate is derived (mainwindow.cpp:1346-1350):

(ChannelSavedCount delta) * 1e9 / (ChannelRealtime delta)

but only when the realtime delta exceeds 1e9 ns and the saved count actually increased. Otherwise it is forced to 0 — so expect genuine zeros on the first tick or two after starting, and on any quiet channel, rather than a gap in the series.

A VX2730 quirk sits immediately above it: ChannelRealtime is divided by 4 because the register is not in ns on that model.

Setup preconditions

SetupInflux() (mainwindow.cpp:2460) refuses to connect unless all of the following hold. If any fails, the influx object is deleted and nothing is ever pushed:

• Both DatabaseIP and DatabaseName are non-empty.
• TestingConnection() succeeds.
• If the server reports version > 1, a token is set. The configuration here is apiv3_… against localhost:8181, so this branch is live and the token is mandatory.
• The database already exists. It is matched against the list returned by CheckDatabases(). Despite CreateDatabase() existing in ClassInfluxDB, it is never called — MUSIC has to be there beforehand.

One latent inconsistency, which cannot bite with current hardware: the register-read loop clamps to MaxNumberOfChannel (64), but the InfluxDB loop at mainwindow.cpp:1371 iterates GetNChannels() unclamped. Every x2730-series board is 64 channels or fewer, so the two agree today; a wider board would read past the haha[] array.