script: update

scripts-r/viz-campaign.R

@@ -35,6 +35,8 @@ data_z_joinable = data_z %>% transmute(
 data_nz = inner_join(data_nz, data_z_joinable) %>% mutate(
   mean_turnaround_time_minus_zero = mean_turnaround_time - zero_mean_turnaround_time,
   mean_waiting_time_minus_zero = mean_waiting_time - zero_mean_waiting_time
+) %>% mutate(
+  mean_turnaround_time_increase_ratio = mean_turnaround_time_minus_zero / zero_mean_turnaround_time
 )
 
 # energy diff from powercap, depending on powercap ratio and predictor
@@ -209,8 +211,40 @@ data_nz %>%
 scale=0.9
 ggsave(sprintf("%s/sched-mtt-distribution.pdf", output_dir), width=8*scale, height=4*scale)
 
-# enable comparison of mean power values between predictors for all (workload, powercap) tuples
+
+# show all columns...
 options(dplyr.width = Inf)
+
+# compute overall power under-utilization compared to the powercap
+data_nz %>%
+  mutate(power_underutilization_ratio = (powercap_dynamic_watts - mean_power)/powercap_dynamic_watts) %>%
+  group_by(predictor_name) %>%
+  summarize(
+    min_power_underutilization_ratio = min(power_underutilization_ratio),
+    average_power_underutilization_ratio = mean(power_underutilization_ratio)
+  )
+
+# compute overall mean turnaround time increase
+data_nz %>%
+  filter(start_dt_s != outlier_workload_start_dt_s) %>%
+  group_by(predictor_name) %>%
+  summarize(
+    average_mtt_increase = mean(mean_turnaround_time_minus_zero),
+    average_mtt_increase_ratio = mean(mean_turnaround_time_increase_ratio)
+  )
+
+# compute overall powercap breaks
+data_nz %>%
+  mutate(powercap_break = pmax(max_power_from_powercap, 0)) %>%
+  mutate(powercap_break_ratio = powercap_break / powercap_dynamic_watts) %>%
+  group_by(predictor_name) %>%
+  summarize(
+    mean_max_power_from_powercap = mean(max_power_from_powercap),
+    min_max_power_from_powercap = min(max_power_from_powercap),
+    max_max_power_from_powercap = max(max_power_from_powercap),
+  )
+
+# enable comparison of mean power values between predictors for all (workload, powercap) tuples
 data_p_mean = data %>% pivot_wider(names_from = predictor_name, values_from = mean_power) %>%
   replace_na(list(max=0,upper_bound=0,zero=0,real_max=0,mean=0,real_mean=0))
 
@@ -259,8 +293,6 @@ against_mean = data_p_mean %>%
 instances_where_some_max_power_is_NOT_below_mean = against_mean %>% filter(!all_max_below_mean)
 print(sprintf("number of occurrences where mean consumes more power than max/real_max: %d/%d", nrow(instances_where_some_max_power_is_NOT_below_mean), nrow(against_mean)))
 
-
-
 data_p_mtt = data_nz %>% filter(start_dt_s != outlier_workload_start_dt_s) %>%
   pivot_wider(names_from = predictor_name, values_from = mean_turnaround_time_minus_zero) %>%
   replace_na(list(max=0,upper_bound=0,zero=0,real_max=0,mean=0,real_mean=0))
@@ -339,6 +371,3 @@ data %>% ggplot() +
     x="Power predictor",
     y="Distribution of the unused energy consumed during the constrained period for each simulation (GJ).\nComputed as the integral of the dynamic power minus the dynamic powercap value, only keeping negative values."
   )
-
-
-