Re: [Ganglia-developers] [PATCH] Weird system/idle cpu metrics (ppc64)

Thu, 06 May 2010 05:40:17 -0700

Hi Jesse,

Here it is.

Thank you so much!

Rafael Xavier de Souza
Linux Technology Center Software Engineer
IBM Systems & Technology Group
[email protected]
MM17 Hortolândia-SP, Brazil

Em 05-05-2010 19:33, Jesse Becker escreveu:
Could you please re-base this patch off of trunk?  Once done, I'll test and commit it.

Thanks for the patch!

On Wed, May 5, 2010 at 09:47, Rafael Xavier de Souza <[email protected]> wrote:
This patch fixes CPU system and idle metrics. Ganglia uses float variables (double) to store the jiffies and the jiffies sums. But, Float numbers have a problem with precision and Ganglia is getting lost with big ppc64 numbers on its calculations. This patch changes libmetrics/linux/metrics.c to use integers (unsined long long) to store the jiffies and jiffies sums and floats (double) to store the calculated numbers only.


Signed off by: Rafael Xavier <[email protected]>

OBS:
1) This patch has been sniff tested, but not extensively.
2) It fixes it for linux only. The ideal approach would be to make the same changes for other OSes as well.

--

Rafael Xavier de Souza
Linux Technology Center Software Engineer
IBM Systems & Technology Group
[email protected]
MM17 Hortolândia-SP, Brazil

------------------------------------------------------------------------------

_______________________________________________
Ganglia-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-developers




--
Jesse Becker
Every cloud has a silver lining, except for the mushroom-shaped ones, which come lined with strontium-90.

Index: libmetrics/linux/metrics.c
===================================================================
--- libmetrics/linux/metrics.c	(revisão 2294)
+++ libmetrics/linux/metrics.c	(cópia de trabalho)
@@ -23,6 +23,8 @@
 #define OSNAME "Linux"
 #define OSNAME_LEN strlen(OSNAME)
 
+#define JT unsigned long long
+
 /* /proc/net/dev hash table stuff */
 typedef struct net_dev_stats net_dev_stats;
 struct net_dev_stats {
@@ -593,11 +595,11 @@
 /*
  * A helper function to return the total number of cpu jiffies
  */
-unsigned long
+JT
 total_jiffies_func ( void )
 {
    char *p;
-   unsigned long user_jiffies, nice_jiffies, system_jiffies, idle_jiffies,
+   JT user_jiffies, nice_jiffies, system_jiffies, idle_jiffies,
                  wio_jiffies, irq_jiffies, sirq_jiffies;
 
    p = update_file(&proc_stat);
@@ -632,8 +634,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_user_jiffies,  user_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_user_jiffies,  user_jiffies, 
+             last_total_jiffies, total_jiffies, diff;
 
    p = update_file(&proc_stat);
    if((proc_stat.last_read.tv_sec != stamp.tv_sec) &&
@@ -647,7 +649,7 @@
      diff = user_jiffies - last_user_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -664,8 +666,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_nice_jiffies,  nice_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_nice_jiffies,  nice_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    p = update_file(&proc_stat);
    if((proc_stat.last_read.tv_sec != stamp.tv_sec) &&
@@ -680,7 +682,7 @@
      diff = (nice_jiffies  - last_nice_jiffies);
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -697,8 +699,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_system_jiffies,  system_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_system_jiffies,  system_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    p = update_file(&proc_stat);
    if((proc_stat.last_read.tv_sec != stamp.tv_sec) &&
@@ -722,7 +724,7 @@
      diff = system_jiffies - last_system_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -739,8 +741,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_idle_jiffies,  idle_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_idle_jiffies,  idle_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    p = update_file(&proc_stat);
    if((proc_stat.last_read.tv_sec != stamp.tv_sec) &&
@@ -757,7 +759,7 @@
      diff = idle_jiffies - last_idle_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -774,7 +776,7 @@
 {
    char *p;
    g_val_t val;
-   double idle_jiffies, total_jiffies;
+   JT idle_jiffies, total_jiffies;
 
    p = update_file(&proc_stat);
 
@@ -782,10 +784,10 @@
    p = skip_token(p);
    p = skip_token(p);
    p = skip_token(p);
-   idle_jiffies  = strtod( p , (char **)NULL );
+   idle_jiffies  = (JT) strtod( p , (char **)NULL );
    total_jiffies = total_jiffies_func();
 
-   val.f = (idle_jiffies/total_jiffies) * 100;
+   val.f = ((double)(idle_jiffies/total_jiffies)) * 100.0;
    return val;
 }
 
@@ -795,8 +797,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_wio_jiffies,  wio_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_wio_jiffies,  wio_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    if (num_cpustates == NUM_CPUSTATES_24X) {
      val.f = 0.0;
@@ -819,7 +821,7 @@
      diff = wio_jiffies - last_wio_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -837,8 +839,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_intr_jiffies,  intr_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_intr_jiffies,  intr_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    if (num_cpustates == NUM_CPUSTATES_24X) {
      val.f = 0.;
@@ -862,7 +864,7 @@
      diff = intr_jiffies - last_intr_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 
@@ -880,8 +882,8 @@
    char *p;
    static g_val_t val;
    static struct timeval stamp={0, 0};
-   static double last_sintr_jiffies,  sintr_jiffies,
-                 last_total_jiffies, total_jiffies, diff;
+   static JT last_sintr_jiffies,  sintr_jiffies,
+             last_total_jiffies, total_jiffies, diff;
 
    if (num_cpustates == NUM_CPUSTATES_24X) {
      val.f = 0.;
@@ -906,7 +908,7 @@
      diff = sintr_jiffies - last_sintr_jiffies;
 
      if ( diff )
-       val.f = (diff/(total_jiffies - last_total_jiffies)) * 100;
+       val.f = ((double)diff/(double)(total_jiffies - last_total_jiffies)) * 100.0;
      else
        val.f = 0.0;
 


------------------------------------------------------------------------------

_______________________________________________
Ganglia-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-developers

Reply via email to