Systemtap vs DTrace comparison

	Systemtap	DTrace
License	GPL	CDDL
Operating System Support	Linux	Solaris, FreeBSD, OSX, netBSD (is planned)
Processor Support	Wide (as per kprobes: x86, x86_64, ppc64, ia64, s390, sparc?, ...)	X86, x86_64, ultrasparc, ppc?
kernel coupling ("upstream" lock-step)	Little	Not an issue for Solaris (built in, full support) Lot
Core Developer s	IBM, RedHat. Some community support	Sun, Apple, Freebsd, community support
Development began	2005	2001
Ongoing Development	Rapid	Mature, normal rate
Target audience	Development. Users, sysadmins	Kernel developers, userland developers, sysadmins, users
Actual audience	Kernel Developers	Kernel developers, userland developers, sysadmins, users
Target usage	Debugging, tracing, profiling	Debugging, tracing, profiling
Language
Style	Scripting	Scripting
Full control structures	Yes	No ((conditionals using ? : )
Variable typing	Implicit, infered	Implicit
Complex report generation	Yes	Yes
Scalable aggregates	Yes	Yes
Thread-local variables	Yes (from first principles via tid-indexed auxiliary arrays)	Yes
Speculative tracing	made redundant by available control structures	Yes
Binary tracing (sends binary formatted packets from kernel to userland	Yes	? (unimportant to the user)
early boot tracing	Not yet	Yes
Probe execution	Optimized native code	Interpreted bytecodes
probing capability
Number of possible probe points	Unlimited (currently limited to every line in the kernel, no userland probles or static probe points by default)	Unlimited (50,000 kernel and static probe points, probe any line of userspace code)
Maximum tested probing points	50,000 active probe test fails	500,000 active probe test works as expected.
probe arbitrary points in code (function entry, exit, interior)	Yes ( currently limited to kernel space)	Yes (kernel limited to static probe points and kernel function entry and exit, userland can probe any line of code)
probe dynamically loaded kernel objects	Yes	Yes
Probe dynamically loaded userland libraries	N/A ( no userland probes)	Yes
probe user-space programs	Soon	Yes
statically inserted probe points, kernel side	Yes (STP_MARK)	Yes (DTRACE_PROBE)
statically inserted probe points, user side, e.g. interpretive languages	Not yet	Yes (DTRACE_PROBE)
timer-based probing	Yes	Yes
hardware performance counter based probing	Coming soon	Coming soon
Safety
time-limited probe handler execution	Yes (statement counting)	Yes (implied by language restriction)
non-blocking, atomic probe handlers	Yes ( may be over ridden, at the cost of safety)	Yes
space-limited execution	Yes	Yes
division-by-zero protection	Yes	Yes
null pointer dereferencing protection	Yes	Yes
means available to bypass protection for advanced users	yes (guru mode, embedded-C)	limited (predefined "destructive" actions available)
translate-time error checks	Many	Many
run-time error checks performed by	automatically generated C code	Byte code interpreter
use by unprivileged users	for now, sudo all-or-nothing	graduated access by privilege level
Actual Usage:
Number of active users	hundreds	10's of thousands
Tested in enterprise production enviroment	Extermely limited	Fully tested and trusted
Number of real world problems solved	few	1000's