improve pm2 plotter:

- error bars
- hot vs. hot error
- keep aggregate CSVs around.

1 files changed, 44 insertions, 14 deletions

diff --git a/plot_pm2.py b/plot_pm2.py
index 09b2d49..40e6a63 100755
--- a/plot_pm2.py
+++ b/plot_pm2.py
@@ -13,7 +13,7 @@ import stats
 import defapp
 
 from plot     import decode
-from gnuplot  import gnuplot, FORMATS
+from gnuplot  import gnuplot, FileGraph, FORMATS
 
 
 
@@ -75,6 +75,8 @@ PMO_AGGR_SUBPLOTS = [
 #    (0, 10,    7,  True),
 #    (0, 10,    8,  True),
     (0, 10,    9,  True),
+    (0,  8,    7,  False), # difference of second to first hot access
+    (0,  9,    8,  False), # difference of third to second hot access
 ]
 
 PMO_AGGR_COMBINE = [
@@ -104,6 +106,7 @@ options = [
     o(None, '--wide', action='store_true', dest='wide'),
     o(None, '--split', action='store_true', dest='split'),
     o(None, '--log-y', action='store_true', dest='logy'),
+    o(None, '--errorbar', action='store_true', dest='errbar'),
     o(None, '--extend', action='store', type='float', dest='extend'),
     o(None, '--aggregate', action='store_true', dest='aggregate'),
     o('-c', '--cycles-per-usec', action='store', type='float', dest='cycles_per_usec'),
@@ -117,7 +120,8 @@ defaults = {
     'aggregate' : False,
     'extend' : 1.5,
     'cycles_per_usec' : None,
-    'logy' : False,
+    'logy'   : False,
+    'errbar' : False,
     }
 
 def extract_cols(data, xcol, ycol1, ycol2, cast=int, cpu_filter=lambda x, y: True):
@@ -346,11 +350,16 @@ class CyclePlotter(defapp.App):
         mems = [tag for (tag, _) in MACHINE_TOPOLOGY[host][1]]
         columns = []
         idx = 2
+        header = ["wss"]
         for (x, y, yminus, split) in PMO_AGGR_SUBPLOTS:
             tags = ['all']
             if split:
                 tags += mems
             for tag in tags:
+                col_name = "%s %s" % (PMO_COL_LABEL[y][1], tag)
+                if not yminus is None:
+                    col_name += ' - ' + PMO_COL_LABEL[yminus][1]
+                header += [col_name + " avg", col_name + " std", col_name + " wc"]
                 columns.append((x, y, yminus, tag, idx))
                 idx += 3
 
@@ -360,7 +369,10 @@ class CyclePlotter(defapp.App):
             data[:, 1:] /= self.options.cycles_per_usec
         else:
             yunit = "(cycles)"
-        tmp = write_csv_file(None, data)
+
+        csvfile = "xxx-%s" % datafile
+
+        write_csv_file(csvfile, data, header, width=max([len(h) for h in header]))
 
         rw = int(conf['wcycle'])
         rw = 1.0 / rw * 100 if rw != 0 else 0
@@ -378,9 +390,10 @@ class CyclePlotter(defapp.App):
                     label = PMO_COL_LABEL[y][0]
                     if y == 10:
                         label += " from %s" % PMO_MEM[tag]
-                    graphs += [
-                        (tmp.name, 1, idx + offset, label),
-                        ]
+                    graphs.append(
+                        FileGraph(
+                            csvfile, xcol=1, ycol=idx + offset, title=label,
+                            error=idx + offset + 1 if kind == 'avg' and self.options.errbar else None))
                 xlabel = "working set size (kilobytes)"
                 ylabel = "time to complete access " + yunit
                 title = "measured %s WSS access time (%.2f%% writes)" % (long, rw)
@@ -397,9 +410,10 @@ class CyclePlotter(defapp.App):
             for (x, y, yminus, tag, idx) in columns:
                 if not (yminus is None) and tag != 'all':
                     label = "%s" % PMO_MEM[tag]
-                    graphs += [
-                        (tmp.name, 1, idx + offset, label),
-                        ]
+                    graphs.append(
+                        FileGraph(
+                            csvfile, xcol=1, ycol=idx + offset, title=label,
+                            error=idx + offset + 1 if kind == 'avg' and self.options.errbar else None))
             xlabel = "working set size (kilobytes)"
             ylabel = "per-sample delta to hot access " + yunit
             title = "measured %s overhead (%.2f%% writes)"  % (long, rw)
@@ -408,7 +422,25 @@ class CyclePlotter(defapp.App):
             fname = "%s_delta_%s" % (name, kind)
             gnuplot(graphs, xlabel=xlabel, ylabel=ylabel, title=title, fname=fname,
                     yrange=yrange, logscale=axis, format=self.options.format)
-        del tmp
+
+            graphs = []
+            for (x, y, yminus, tag, idx) in columns:
+                if y in [8, 9] and yminus in [7, 8] and tag == 'all':
+                    label = "%s to %s" % (PMO_COL_LABEL[yminus][0], PMO_COL_LABEL[y][0])
+                    graphs.append(
+                        FileGraph(
+                            csvfile, xcol=1, ycol=idx + offset, title=label,
+                            error=idx + offset + 1 if kind == 'avg' and self.options.errbar else None))
+            xlabel = "working set size (kilobytes)"
+            ylabel = "per-sample delta to previous hot access " + yunit
+            title = "measured %s differences (%.2f%% writes)"  % (long, rw)
+            yrange = None
+
+            fname = "%s_delta-h_%s" % (name, kind)
+            gnuplot(graphs, xlabel=xlabel, ylabel=ylabel, title=title, fname=fname,
+                    yrange=yrange, logscale=axis, format=self.options.format)
+
+#        del tmp
 
         # stats delta
         # find hot column
@@ -433,9 +465,7 @@ class CyclePlotter(defapp.App):
                 if yminus is None and tag != 'all':
                     label = PMO_COL_LABEL[y][0]
                     label = PMO_MEM[tag]
-                    graphs += [
-                        (tmp.name, 1, idx + offset, label),
-                        ]
+                    graphs.append(FileGraph(tmp.name, xcol=1, ycol=idx+offset, title=label))
                 xlabel = "working set size (kilobytes)"
                 ylabel = "delta to third hot access " + yunit
                 title = "difference of %s access costs (%.2f%% writes)" % (long, rw)
@@ -444,7 +474,7 @@ class CyclePlotter(defapp.App):
                 fname = "%s_diff_%s" % (name, kind)
                 gnuplot(graphs, xlabel=xlabel, ylabel=ylabel, title=title, fname=fname,
                         yrange=yrange, logscale=axis, format=self.options.format)
-        del tmp
+#        del tmp
 
     def plot_file(self, datafile):
         bname = basename(datafile)