Work on a simple workload generator for benchmarking.

diff --git a/microbench/workload.py b/microbench/workload.py
new file mode 100755 (executable)
index 0000000..d459b9d
--- /dev/null
+++ b/microbench/workload.py
@@ -0,0 +1,87 @@
+#!/usr/bin/python
+#
+# A simple benchmark for Blue Sky that will read and/or write a collection of
+# files with a specified working set size, and measure the response time to do
+# so.
+
+import json, os, random, sys, threading, time
+
+THREADS = 1
+
+class WorkerThread:
+    """Performs a mix of file system operations and records the performance."""
+
+    def __init__(self):
+        self.stats = []
+        self.duration = 10.0        # Seconds for which to run
+        self.write_fraction = 0.3   # Fraction of operations which are writes
+        self.wss_count = 16         # Files in the working set
+        self.tot_count = 32         # Total number of files created
+        self.filesize = 32 * 1024   # Size of files to work with
+        self.target_ops = 2         # Target operations/second/thread
+
+    def setup(self):
+        for i in range(self.tot_count):
+            filename = "file-%d" % (i,)
+            fp = open(filename, 'w')
+            fp.write('\0' * self.filesize)
+            fp.close()
+
+    def run(self):
+        stop_time = time.time() + self.duration
+
+        while True:
+            time1 = time.time()
+            if time1 >= stop_time: break
+            info = self._operation()
+            time2 = time.time()
+            self.stats.append((time1, time2 - time1, info))
+            print self.stats[-1]
+            delay = time1 + (1.0 / self.target_ops) - time2
+            if delay > 0: time.sleep(delay)
+
+    def _operation(self):
+        """Run a single file system test (i.e., read or write a file)."""
+
+        filename = "file-%d" % (random.randrange(self.wss_count),)
+
+        if random.uniform(0.0, 1.0) < self.write_fraction:
+            fp = open(filename, 'w')
+            fp.write('\0' * self.filesize)
+            fp.close()
+            return ('write', filename)
+        else:
+            fp = open(filename, 'r')
+            fp.read()
+            fp.close()
+            return ('read', filename)
+
+def run_stats(stats):
+    duration = max(x[0] for x in stats) - min(x[0] for x in stats)
+    latencies = [x[1] for x in stats]
+    latencies.sort()
+    print "Experiment duration:", duration
+    print "Operation count:", len(stats)
+    print "Latencies:", latencies
+    print "Average latency:", sum(latencies) / len(latencies)
+
+if __name__ == '__main__':
+    workers = []
+    threads = []
+    for i in range(THREADS):
+        w = WorkerThread()
+        if i == 0: w.setup()
+        t = threading.Thread(target=w.run)
+        threads.append(t)
+        workers.append(w)
+        t.start()
+    for t in threads:
+        t.join()
+
+    results = []
+    for w in workers:
+        results += w.stats
+    results.sort()
+
+    print json.dumps(results, indent=2)
+    run_stats(results)