<?xml version="1.0" encoding="utf-8"?>
<feed xml:lang="en" xmlns="http://www.w3.org/2005/Atom"><title>Recent changes to support-requests</title><link href="https://sourceforge.net/p/lmbench/support-requests/" rel="alternate"/><link href="https://sourceforge.net/p/lmbench/support-requests/feed.atom" rel="self"/><id>https://sourceforge.net/p/lmbench/support-requests/</id><updated>2006-09-18T08:01:44Z</updated><subtitle>Recent changes to support-requests</subtitle><entry><title>run signal tests in DEVELOPMENT mode</title><link href="https://sourceforge.net/p/lmbench/support-requests/2/" rel="alternate"/><published>2006-09-18T08:01:44Z</published><updated>2006-09-18T08:01:44Z</updated><author><name>Anonymous</name><uri>https://sourceforge.net/u/userid-None/</uri></author><id>https://sourceforge.nete506573d30993602fa2ebd1944082cc3838f3ebb</id><summary type="html">&lt;div class="markdown_content"&gt;&lt;p&gt;Hi,&lt;br /&gt;
We are using lmbench  in Linux Kernel &lt;br /&gt;
Performance project &lt;a href="http://kernel-" rel="nofollow"&gt;http://kernel-&lt;/a&gt;&lt;br /&gt;
perf.sourceforge.net. To get results for other than &lt;br /&gt;
default lmbench CPU scheduler and to narrow &lt;br /&gt;
performance tool applying the DEVELOPMENT mode was &lt;br /&gt;
used instead of OS mode in ‘make result’. But signal &lt;br /&gt;
test is not running in DEVELOPMENT mode because &lt;br /&gt;
BENCHMARK_SIG is not set in config-run.&lt;br /&gt;
To fix the problem with minimal modifications &lt;br /&gt;
and saving ‘make results’ interface next way was &lt;br /&gt;
chosen:&lt;br /&gt;
--- lmbench-3.0-a7/scripts//lmbench     2005-&lt;br /&gt;
03-30 01:54:05.000000000 -0800&lt;br /&gt;
+++ lmbench-3.0-a7m/scripts//lmbench    2006-09-15 &lt;br /&gt;
04:17:39.000000000 -0700&lt;br /&gt;
@@ -153,7 +153,7 @@ if [ X$BENCHMARK_OS = XYES -o &lt;br /&gt;
X$BENCHMAR&lt;br /&gt;
do      lat_select -n $i -P $SYNC_MAX tcp&lt;br /&gt;
done&lt;br /&gt;
fi&lt;br /&gt;
-if [ X$BENCHMARK_OS = XYES -o X$BENCHMARK_SIG = &lt;br /&gt;
XYES ]; then&lt;br /&gt;
+if [ X$BENCHMARK_OS = XYES -o X$BENCHMARK_PROC = &lt;br /&gt;
XYES ]; then&lt;br /&gt;
lat_sig -P $SYNC_MAX install&lt;br /&gt;
lat_sig -P $SYNC_MAX catch&lt;br /&gt;
lat_sig -P $SYNC_MAX prot lat_sig&lt;/p&gt;
&lt;p&gt;Leonid&lt;/p&gt;&lt;/div&gt;</summary></entry><entry><title>as far as possible CPU culculating fix</title><link href="https://sourceforge.net/p/lmbench/support-requests/1/" rel="alternate"/><published>2006-09-18T07:21:57Z</published><updated>2006-09-18T07:21:57Z</updated><author><name>Anonymous</name><uri>https://sourceforge.net/u/userid-None/</uri></author><id>https://sourceforge.net513e869ca06904beb08074a9a7dcafc43f776bd2</id><summary type="html">&lt;div class="markdown_content"&gt;&lt;p&gt;Hi,&lt;/p&gt;
&lt;p&gt;lmbench-3.0-a7 has a new useful option to control job &lt;br /&gt;
placement. The option #5 (spread) says:&lt;br /&gt;
5) Assign each benchmark and attendant processes to &lt;br /&gt;
their own processors, except that they will be as far &lt;br /&gt;
as possible from each other and other processes&lt;/p&gt;
&lt;p&gt;But actually the same CPU number is sorted by &lt;br /&gt;
reverse_bits() for CPU numbers 0,2,5,7 as you can see &lt;br /&gt;
in the next table&lt;/p&gt;
&lt;p&gt;Prev_cpu  Next_cpu&lt;br /&gt;
0        0&lt;br /&gt;
1        4&lt;br /&gt;
2        2&lt;br /&gt;
3        6&lt;br /&gt;
4        1&lt;br /&gt;
5        5&lt;br /&gt;
6        3&lt;br /&gt;
7        7&lt;/p&gt;
&lt;p&gt;Furthermore function reverse_bits() does not take &lt;br /&gt;
into account CPU physical package id while calculates &lt;br /&gt;
as far as possible CPU.&lt;/p&gt;
&lt;p&gt;Proposed patch calculates next cpu number on the &lt;br /&gt;
basis of files&lt;br /&gt;
/sys/devices/system/cpu/cpuNNN/topology/physical_p&lt;br /&gt;
ackage_id&lt;br /&gt;
if the file is present. To avoid looking into the &lt;br /&gt;
file during benchmark run time the array &lt;br /&gt;
cpu_other_package[] number is created by &lt;br /&gt;
lmbench/scripts/build in the file lib_sched.h at a &lt;br /&gt;
time of  lmbench install. So it is not need to read &lt;br /&gt;
physical_package_id inside measured by benchmark time &lt;br /&gt;
interval.&lt;br /&gt;
Next algorithm is used for as far as possible &lt;br /&gt;
CPU number calculating&lt;br /&gt;
do&lt;br /&gt;
next_cpu=(next_cpu++)%num_of_cpu;&lt;br /&gt;
untile (pack_id [prev_cpu] != pack_id [next_cpu])&lt;/p&gt;
&lt;p&gt;If there is no file physical_package_id in the system &lt;br /&gt;
or all id’s are the same, next CPU is calculated &lt;br /&gt;
using formula &lt;br /&gt;
(cpu + num_of_cpus / 2 )  % num_of_cpus&lt;br /&gt;
because as far as possible CPU is a diameter opposite &lt;br /&gt;
CPU if we consider CPUs set as a ring.&lt;/p&gt;
&lt;p&gt;Leonid&lt;/p&gt;&lt;/div&gt;</summary></entry></feed>