Re: MYSQL,Linux large threads number
On Fri, 15 Mar 2002, Peter Zaitsev wrote: Hello mysql, Some time ago I wrote about slow down of mysql with large number of threads, which is quite common in Linux-Apache-Mysql-PHP enviroment. The test was simulating the worst case of concurrency - all the threads are modified global variable in a loop 500 times in total, using standard mutex for synchronization. The yeild is used in a loop to force even more fair distribution of lock usage by threads and increase context switches, therefore it did not change much with large number of threads. I.e with 64 threads time without yeild is 1:33.5 For Next test I'll try to use Ingo's scheduler to see if it helps to solve the problem, also I'll try to test real mysql server to see which slowdown it will have. I think it would be instructive to try using NGPT and either a patched kernel or 2.4.19-pre3 which has the extra system calls included. I make no predictions, but large improvements have been noted with a large number of threads. CODE: (Dumb one just for test) #include stdio.h #include pthread.h #include time.h #define NUM_TH 1000 You might want to read this from the command line as an option and malloc the vecors. Just a thought, I'm bad at guessing how lare a test I may want to run. -- bill davidsen [EMAIL PROTECTED] CTO, TMR Associates, Inc Doing interesting things with little computers since 1979. - Before posting, please check: http://www.mysql.com/manual.php (the manual) http://lists.mysql.com/ (the list archive) To request this thread, e-mail [EMAIL PROTECTED] To unsubscribe, e-mail [EMAIL PROTECTED] Trouble unsubscribing? Try: http://lists.mysql.com/php/unsubscribe.php
MYSQL,Linux large threads number
Hello mysql,
Some time ago I wrote about slow down of mysql with large number of
threads, which is quite common in Linux-Apache-Mysql-PHP enviroment.
The test was simulating the worst case of concurrency - all the
threads are modified global variable in a loop 500 times in
total, using standard mutex for synchronization. The yeild is used
in a loop to force even more fair distribution of lock usage by
threads and increase context switches, therefore it did not change
much with large number of threads. I.e with 64 threads time without
yeild is 1:33.5
Test was run on PIII-500 1G RAM Kernel 2.4.18. 3 runs were made for
each number of threads and best results were taken:
Num threads. Time Peak cs rate.
2 53.4 179518
4 53.8 144828
16 1:06.3 85172
64 1:48.1 48394
256 8:10.6 10235
1000 36:46.2 2602
The surprising thing is the time grows in less then linear way for up
to 64 threads but later it stars to go linear way or even worse. May
be this is because some other process are sleeping in the system which
also is used in scheduling.
For Next test I'll try to use Ingo's scheduler to see if it helps to
solve the problem, also I'll try to test real mysql server to see
which slowdown it will have.
CODE: (Dumb one just for test)
#include stdio.h
#include pthread.h
#include time.h
#define NUM_TH 1000
#define TOTAL_VALUE 500
#define LOOP (TOTAL_VALUE/NUM_TH)
pthread_t th[NUM_TH];
int thread_data[NUM_TH];
int rc,rc2;
int global=0;
pthread_mutex_t mut = PTHREAD_MUTEX_INITIALIZER;
pthread_mutex_t start = PTHREAD_MUTEX_INITIALIZER;
void f(int *thn)
{
int i;
pthread_mutex_lock(start);
pthread_mutex_unlock(start);
for (i=0;iLOOP;i++)
{
pthread_mutex_lock(mut);
global++;
pthread_yield();
pthread_mutex_unlock(mut);
}
}
main()
{
int i;
pthread_mutex_lock(start);
for (i=0;iNUM_TH;i++)
{
thread_data[i]=i;
rc=pthread_create(th[i],NULL,f,thread_data[i]);
if (rc!=0)
{
printf(Failed to create thread #%d errorcode:%d\n,i,rc);
}
}
pthread_mutex_unlock(start);
for (i=0;iNUM_TH;i++)
{
rc2=pthread_join(th[i],NULL);
}
printf(Global Value: %d\n,global);
}
--
Best regards,
Peter mailto:[EMAIL PROTECTED]
-
Before posting, please check:
http://www.mysql.com/manual.php (the manual)
http://lists.mysql.com/ (the list archive)
To request this thread, e-mail [EMAIL PROTECTED]
To unsubscribe, e-mail [EMAIL PROTECTED]
Trouble unsubscribing? Try: http://lists.mysql.com/php/unsubscribe.php
Re: MYSQL,Linux large threads number
Peter Zaitsev writes: Hello mysql, Some time ago I wrote about slow down of mysql with large number of threads, which is quite common in Linux-Apache-Mysql-PHP enviroment. The test was simulating the worst case of concurrency - all the threads are modified global variable in a loop 500 times in total, using standard mutex for synchronization. The yeild is used in a loop to force even more fair distribution of lock usage by threads and increase context switches, therefore it did not change much with large number of threads. I.e with 64 threads time without yeild is 1:33.5 Test was run on PIII-500 1G RAM Kernel 2.4.18. 3 runs were made for each number of threads and best results were taken: Num threads. Time Peak cs rate. 2 53.4 179518 4 53.8 144828 16 1:06.3 85172 64 1:48.1 48394 256 8:10.6 10235 1000 36:46.2 2602 I hope you are using latest glibc 2.2 with our patches applied. Also please try adaptive mutex. How to set it, take a look at mysys/my_thr_init.c. -- Regards, __ ___ ___ __ / |/ /_ __/ __/ __ \/ /Mr. Sinisa Milivojevic [EMAIL PROTECTED] / /|_/ / // /\ \/ /_/ / /__ MySQL AB, Fulltime Developer /_/ /_/\_, /___/\___\_\___/ Larnaca, Cyprus ___/ www.mysql.com - Before posting, please check: http://www.mysql.com/manual.php (the manual) http://lists.mysql.com/ (the list archive) To request this thread, e-mail [EMAIL PROTECTED] To unsubscribe, e-mail [EMAIL PROTECTED] Trouble unsubscribing? Try: http://lists.mysql.com/php/unsubscribe.php
