A Solaris users guide to Linux
Phil Kirk
Introduction
What this presentation is
A look at what equivalent features exist in Linux and how theyre used Features that I use or find interesting
What this presentation isnt
Which operating system is better than the other How to setup a linux machine
Environments arent just Solaris or Linux, theres usually a different way to achieve the same thing
Userland tools
Truss Still used a lot even with newer tools like DTrace available Common options, -f, -e, -a sotruss or truss u trace library calls
Userland tools cont.
Example
Userland tools cont.
strace Much like truss, very similar options ltrace trace dynamic libraries as well as system calls But there are differences No stop on signal, depending on ltrace version problems with threads Example
Userland tools cont.
Linkers and compilers
Lots of similar linker options LD_DEBUG, LD_PRELOAD, LD_LIBRARY_PATH man ld.so, ld.so.1 Examples where the linker can be exploited to help find problems LD_DEBUG, which libraries are getting loaded in what order LD_PRELOAD, interpose on library calls
Linkers and compilers cont.
gcc and cc have the same set of options they just have different names Same for gdb and dbx Of course gcc and gdb run on Solaris as well Useful gdb things, gdb scripts and gdb tty mode , though with ddd or other graphical debuggers perhaps you dont need to
PTools
pstack, pgrep, pldd, etc. Exist on both Solaris and Linux but some things arent quite the same and Linux doesnt have all the same tools As an example pstack under linux doesnt work on core files Of course its its easy enough to use gdb but...
Solaris DTrace
Everyones familar with Dtrace now but just incase Gives observability into both the kernel and userland Scriptable via D, an awk like language minus looping and conditional structures to prevent unsafe operations Production safe
Solaris Dtrace cont.
Things like the Dtrace tool kit make it very easy to answer questions about whats happening on the system Examples
System tap
Not always part of the installed distribution Gives observability into both the kernel and userland though for userland you may need debuginfo packages Stap scripts are more C like and allow looping constructs, safety is provided inside the SystemTap framework Production system safe?
SystemTap cont.
Lots of examples much like the Dtrace tool kit to help answer common questions quickly As a comparison I noticed there wasnt an opensnoop.stp in the examples so wondered how hard it would be to write Examples
Perf
Originally developed as a userland program to provide access to the prerformance counters Extended to work with kernel tracepoints Can be used to look at userland too http://www.linuxkongress.org/2010/slides/lk2010-perfacme.pdf Examples
LTT
Linux Trace Toolkit Aims to produce an effecient full system tracing acility Allow tracing userland, kernel and also provides tools for viewing and analysing trace output http://lttng.org/
DTrace
Paul Foxs port OLE
Kernel debugging
On Solaris mdb, kmdb Crash dump facilities are managed through dumpadm(1M) All installed as part of core system
Linux tools
kgdb Kdb crash Crash dump facilities managed via kexec/kdump On SMP machines extra config needed Tools not part of many standard distributions and additional packages need to be installed
Kernel debugging cont.
Different approaches to debugging tools Linux tools tend to aim and kernel developers, source code debugging via kgdb Solaris tools developed to root cause system problems the first time they occur
Kernel debugging cont.
What sort of problems do you want to solve? As an example consider a system falling over due to a read/write lock not been released On Solaris its fairly straightforward to start trying to find the culprit First find the lock were interested in
mdb>::walk thread | ::findstack
Then find who owns the lock
Kernel debugging cont
mdb><addr>::kgrep | ::whatis
On Linux it wouldnt be quite so straightforward, crash doesnt have the same feature set as Solaris.
Virtualization
Solaris zones Open Solaris, KVM LDOMS
Solaris zones
Resource containers Lightweight virtualization Configured using zoneadm(1M) zonecfg(1M) Rich feature set Example
Solaris Zones cont.
Easy to create a new zone
#zonecfg z zone testzone: No such zone configured Use 'create' to begin configuring a new zone. zonecfg:testzone> create
Easy to add network interface using vnics
#dladm
create-vnic l e1000g0 vnic0
Easy to administer via zoneadm
Linux OpenVZ
Not part of the core distribution Mature product used as the base for Parallels Virtuozzo Container Looks fairly straightforward to setup containers http://wiki.openvz.org/Main_Page
Linux Vserver
Not part of the standard distribution Need to install separate packages and kernel updates http://linux-vserver.org/Welcome_to_LinuxVServer.org
Linux containers lxc
http://lxc.sourceforge.net/ LXC is the userspace control package for Linux Containers, a lightweight virtual system mechanism sometimes described as chroot on steroids. Mainstream, integrated from 2.6.29 Seems very similar to zones in terms of configuration Not yet as feature rich
Linux containers lxc cont
Uses libvirt to provide virtualized networking support Example of creating a lxc container
Various templates available to create linux guests Creating a new fedora guest is as easy as running lxc-fedora though you may want to do more Running the new guest lxc-start Similar commands to zones
Linux containers lxc cont
lxc-info, lxc-console, lxc-stop
Linux KVM
Type 2 hypervisor requires Intel VT or AMD V cpu support In the linux kernel from 2.6.25 Possibly require additional packages depending on distribution Using virt-manager makes setup and management of virtual machines easy
Linux KVM cont.
Example
Linux KVM cont
Linux KVM cont
Linux KVM cont
Thankyou
Any questions
