System requirements

The system requirements for the Undo Engine are as follows:

  • 64-bit Linux kernel

  • GNU C Library (glibc)

  • GNU POSIX Threads library (libpthread)

The Undo Engine does not require any kernel patches or loaded kernel modules. Most Linux distributions therefore support the Undo Engine without modification. The supported kernel versions are those shipped with the supported distributions listed below.

Furthermore, there is no need to re-build applications; any binary executable can be debugged with the Undo Engine. It is recommended to build executables with debug information, typically by passing the -g option to the compiler, since this allows the debugger to extract more information about the state of the process.

Linux distributions

The Undo Engine is supported on the following Linux distributions:

  • Red Hat Enterprise Linux (and CentOS where applicable) 7.9, 8.6, 8.8, 9.0 1, 9.2 1.

  • Fedora 37 1, 38 1.

  • SUSE Linux Enterprise Server 12 SP5 (12.5), 15 SP4 (15.4), 15 SP5 (15.5).

  • Ubuntu 20.04 LTS, 22.04 1 LTS, 23.04 1.

It will run on most other reasonably modern Linux distributions and versions.

Architectures

The Undo Engine is supported on the following CPU architectures:

  • 32-bit and 64-bit processes running on x86-64

Support for ARMv8 architectures is currently available only to selected customers. Contact Undo Support for more information.

GDB

UDB depends on GDB (currently GDB 13.2) which is supplied as part of the release.

Other operating systems

At this time the Undo Engine does not support non-Linux operating systems such as Windows or macOS. Contact Undo Support if you’re interested in support for a non-Linux operating system.

RAM usage

The main memory requirements for the Undo Engine come from snapshots and the event log.

Regarding snapshots, the Undo Engine normally stores about 30 of these by default, but they’re forked from the original process using copy-on-write so they’re created efficiently. In a short run of the program, these snapshots will be spaced closely together, while in a longer run, they will be spaced further apart. So, the memory consumption for snapshots remains fairly independent of the run time, but reverse-execution commands such as reverse-next and reverse-continue run much faster when there is a snapshot available more recently in the program history. In other words, longer runs affect replay performance rather than snapshot memory utilization.

Regarding the event log, its available space is consumed at a rate that corresponds to the volume of non-deterministic event behaviour. In programs that spend most of their time waiting for something to happen, the memory usage will be low. It’s true that in more non-deterministic applications (for example, those that perform lots of I/O operations) the memory that was allocated to store the event log will sometimes fill up. The size of this memory region is configurable at startup using the --max-event-log-size command-line option, or interactively using the set max-event-log-size command.

Disk usage

During normal operation of LiveRecorder, no information is written to disk: all information is stored in memory. When a recording is saved by LiveRecorder, the in-memory representation of the recorded process is synchronously written to disk as a recording file.

The size of this recording file is determined by various factors. Most notably, the amount of non-deterministic data generated by the process and the size of the process’ memory image. Recording files are compressed, so the size of the recording file is often less than the combined sizes of the memory image and non-deterministic event log.

1(1,2,3,4,5,6)

See limitation regarding Restartable Sequences when attaching to a running program.