map CPU ID to threads to cores to sockets with hyperthreading

Posted on by
Reply

The Linux OS uses a predictable method to map hyperthread enabled CPU threads to CPU IDs. Here’s a reminder if you need a quick way to map them:

# egrep 'processor|core id|physical id' /proc/cpuinfo | cut -d : -f 2 | paste - - -  | awk '{print "CPU"$1"\tsocket "$2" core "$3}'
CPU0    socket 0 core 0
CPU1    socket 0 core 1
CPU2    socket 0 core 2
CPU3    socket 0 core 3
CPU4    socket 1 core 0
CPU5    socket 1 core 1
CPU6    socket 1 core 2
CPU7    socket 1 core 3
CPU8    socket 0 core 0
CPU9    socket 0 core 1
CPU10   socket 0 core 2
CPU11   socket 0 core 3
CPU12   socket 1 core 0
CPU13   socket 1 core 1
CPU14   socket 1 core 2
CPU15   socket 1 core 3

This is a dual quad core system with hyperthreading enabled. 2 physical CPUs with 4 cores each and 2 threads per core so the OS sees 16 CPUs. CPU0 and CPU8 are 2 threads on the first core (core 0 ) on the first physical processor (socket 0)

insert characters into string with sed

Posted on by
Reply 

# echo 20150429 | sed -e 's/\(.\{4\}\)\(.\{2\}\)/\1\/\2\//'
2015/04/29

Start with a single flat directory with thousands of log files…

# ls | head -5
db301.20140216.log.gz
db301.20140217.log.gz
db301.20140218.log.gz
db301.20140219.log.gz
db301.20140220.log.gz

Now move timestamped files into sorted directory by day

# for i in `ls`; do j=$(echo $i| cut -d . -f 2  | sed -e 's/\(.\{4\}\)\(.\{2\}\)/\1\/\2\//'); mkdir -p $j && mv $i $j; done;


check your work

# find . -type f | head -5
./2014/02/16/db301.20140216.log.gz
./2014/02/17/db301.20140217.log.gz
./2014/02/18/db301.20140218.log.gz
./2014/02/19/db301.20140219.log.gz
./2014/02/20/db301.20140220.log.gz

forgot to rename files

# for i in `find . -mindepth 3 -type d `; do pushd $i; for j in `ls`; do k=$(echo $j | sed -e 's/\(\.[0-9]\{8\}\)//' ); mv $j $k;done; popd; done;

check your work

# find . -type f | head -5
./2014/02/16/db301.log.gz
./2014/02/17/db301.log.gz
./2014/02/18/db301.log.gz
./2014/02/19/db301.log.gz
./2014/02/20/db301.log.gz

get list of most recently installed packages with dpkg log

Posted on by
Reply 

# zcat -f /var/log/dpkg.log*|grep 'status installed'|awk '{print $1" "$5}'|sort -u |less
2015-05-04 ffmpeg
2015-05-04 libavcodec52
2015-05-04 libavdevice52
2015-05-04 libavfilter0
2015-05-04 libavformat52
2015-05-04 libavutil49
2015-05-04 libdc1394-22
2015-05-04 libdirac-encoder0
2015-05-04 libdirectfb-1.2-9
2015-05-04 libfaad2
2015-05-04 libgsm1
2015-05-04 libogg0
2015-05-04 libopenjpeg2
2015-05-04 liborc-0.4-0
2015-05-04 libpostproc51
2015-05-04 libraw1394-11
2015-05-04 libschroedinger-1.0-0
2015-05-04 libsdl1.2debian
2015-05-04 libsdl1.2debian-alsa
2015-05-04 libspeex1
2015-05-04 libsvga1
2015-05-04 libswscale0
2015-05-04 libtheora0
2015-05-04 libusb-1.0-0
2015-05-04 libvorbis0a
2015-05-04 libvorbisenc2
2015-05-04 libx86-1
2015-05-04 man-db

IRQBALANCE_BANNED_CPUS explained and working on Ubuntu 14.04

Posted on by
3

Despite the explanation and examples in the manual, there is still some confusion on how to use irqbalance to move interrupts to a specific set of CPUs. (For the sake of simplicity, I use the words CPUs in the same way Linux sees them; it doesn’t care whether they are physical, or cores, or threads.)

Let’s start with a box that has 8 CPUs

# grep processor /proc/cpuinfo 
processor       : 0
processor       : 1
processor       : 2
processor       : 3
processor       : 4
processor       : 5
processor       : 6
processor       : 7

irqbalance uses a variable set at runtime ( IRQBALANCE_BANNED_CPUS ) to make decisions which CPUs should not receive interrupts. IRQBALANCE_BANNED_CPUS is a hexidecimal representation of a “bitmask” to ignore certain CPUs when deciding which CPU to use for a specific interrupt. If we have a box with 8 cpus we will need to start with a binary string 8 digits long: “00000000”

0 means “allow interrupts on this CPU”
1 means “prevent interrupts on this CPU”

In the case of “00000000” we are allowing interrupts on all 8 CPUs. In binary numbers the least significant bit is on the right, mapping the CPUs like this:
“CPU7 CPU6 CPU5 CPU4 CPU3 CPU2 CPU1 CPU0″

Let’s say we want to allow interrupts on the 6th and 7th CPU, and exclude the 1st, 2nd, 3rd, 4th, 5th, and 8th CPU from interrupts. Our bitmask in this case would look like this”

   1    0    0    1    1    1    1    1
"CPU7 CPU6 CPU5 CPU4 CPU3 CPU2 CPU1 CPU0"

This can get tricky. Most humans start counting from 1 but computers start counting at 0. So be careful: the 7th CPU is CPU6 in this case.

Now that we have our bitmask of “10011111” we need to convert it to hexadecimal. You could use this handy bash function:

#
bin2hex() {
   echo "obase=16;ibase=2;$1"|bc
}
# bin2hex 10011111
9F
#

So now we can edit /etc/default/irqbalance and set the variable:

export IRQBALANCE_BANNED_CPUS="9F"

NOTE: I have been able to get irqbalance working on Ubuntu 14.04 trusty only by exporting the IRQBALANCE_BANNED_CPUS variable using “export” in /etc/default/irqbalance AND using the latest (at the time of this writing) irqbalance 1.0.9

Once you understand how the CPUs translate to the bitmask you can extrapolate out from there and come up with a way to ban specific CPUs from irqbalance for any number of CPUs.

ESXi boot device lost connectivity

Posted on by
1

Lost connectivity to the device backing the boot filesystem. As a result, host configuration changes will not be saved to persistent storage.

(slow NetApp filer failover caused lost connectivity to ESXi boot device which is a LUN connected with iSCSI)

# /etc/init.d/hostd restart && /etc/init.d/vpxa restart