Added location parameter to GPareto distribution. Validated value-size distribution against Atikoglu's paper by comparing CDF and PDF.

Generator.cc

@@ -7,7 +7,7 @@
 Generator* createFacebookKey() { return new GEV(30.7984, 8.20449, 0.078688); }
 
 Generator* createFacebookValue() {
-  Generator* g = new GPareto(214.476, 0.348238);
+  Generator* g = new GPareto(15.0, 214.476, 0.348238);
 
   Discrete* d = new Discrete(g);
   d->add(0.00536, 0.0);
@@ -29,7 +29,7 @@ Generator* createFacebookValue() {
   return d;
 }
 
-Generator* createFacebookIA() { return new GPareto(16.0292, 0.154971); }
+Generator* createFacebookIA() { return new GPareto(0, 16.0292, 0.154971); }
 
 Generator* createGenerator(std::string str) {
   if (!strcmp(str.c_str(), "fb_key")) return createFacebookKey();
@@ -68,7 +68,7 @@ Generator* createGenerator(std::string str) {
   if      (strcasestr(str.c_str(), "fixed")) return new Fixed(a1);
   else if (strcasestr(str.c_str(), "normal")) return new Normal(a1, a2);
   else if (strcasestr(str.c_str(), "exponential")) return new Exponential(a1);
-  else if (strcasestr(str.c_str(), "pareto")) return new GPareto(a1, a2);
+  else if (strcasestr(str.c_str(), "pareto")) return new GPareto(a1, a2, a3);
   else if (strcasestr(str.c_str(), "gev")) return new GEV(a1, a2, a3);
   else if (strcasestr(str.c_str(), "uniform")) return new Uniform(a1);
 

Generator.h

@@ -121,23 +121,24 @@ private:
 
 class GPareto : public Generator {
 public:
-  GPareto(double _scale = 1.0, double _shape = 1.0) :
-    scale(_scale), shape(_shape) {
+  GPareto(double _loc = 0.0, double _scale = 1.0, double _shape = 1.0) :
+    loc(_loc), scale(_scale), shape(_shape) {
     assert(shape != 0.0);
-    D("GPareto(scale=%f, shape=%f)", scale, shape);
+    D("GPareto(loc=%f, scale=%f, shape=%f)", loc, scale, shape);
   }
 
   virtual double generate(double U = -1.0) {
     if (U < 0.0) U = drand48();
-    return scale * (pow(U, -shape) - 1) / shape;
+    return loc + scale * (pow(U, -shape) - 1) / shape;
   }
 
   virtual void set_lambda(double lambda) {
     if (lambda <= 0.0) scale = 0.0;
-    else scale = (1 - shape) / lambda;
+    else scale = (1 - shape) / lambda - (1 - shape) * loc;
   }
 
 private:
+  double loc /* mu */;
   double scale /* sigma */, shape /* k */;
 };
 
@@ -177,7 +178,6 @@ public:
     }
 
     return def->generate(Uc);
-    //    return 0.0;
   }
 
   void add(double p, double v) {

TestGenerator.cc

@@ -55,6 +55,12 @@ int main(int argc, char **argv) {
   KeyGenerator kg(g);
   */
 
+  Generator *g = createFacebookValue();
+  //  Generator *g = createGenerator("pareto:15,214.476,0.348238");
+  for (int i = 0; i < 1000000; i++)
+    printf("%f\n", g->generate());
+
+  /*
   Generator *p2 = createGenerator("p:214.476,0.348238");
   //  for (int i = 0; i < 1000; i++)
   //    printf("%f\n", p2->generate());
@@ -62,6 +68,7 @@ int main(int argc, char **argv) {
   p2->set_lambda(1000);
   for (int i = 0; i < 1000; i++)
     printf("%f\n", p2->generate());
+  */
 
   //  for (int i = 0; i < 10000; i++)
   //    printf("%s\n", kg.generate(i).c_str());

cmdline.ggo

@@ -12,7 +12,7 @@ text "\nBasic options:"
 
 option "server" s "Memcached server hostname[:port].  \
 Repeat to specify multiple servers." string multiple
-option "qps" q "Target aggregate QPS." int default="0"
+option "qps" q "Target aggregate QPS. 0 = peak QPS." int default="0"
 option "time" t "Maximum time to run (seconds)." int default="5"
 
 option "keysize" K "Length of memcached keys (distribution)."
@@ -34,8 +34,9 @@ option "depth" d "Maximum depth to pipeline requests." int default="1"
 option "roundrobin" R "Assign threads to servers in round-robin fashion.  \
 By default, each thread connects to every server."
 
-option "iadist" i "Inter-arrival distribution (distribution)."
-       string default="exponential"
+option "iadist" i "Inter-arrival distribution (distribution).  Note: \
+The distribution will automatically be adjusted to match the QPS given \
+by --qps." string default="exponential"
 
 option "noload" - "Skip database loading."
 option "loadonly" - "Load database and then exit."
@@ -62,16 +63,19 @@ Some options take a 'distribution' as an argument.
 Distributions are specified by <distribution>[:<param1>[,...]].
 Parameters are not required.  The following distributions are supported:
 
-   [fixed:]<value>            Always generates <value>.
-   uniform:<max>              Uniform distribution between 0 and <max>.
-   normal:<mean>,<sd>         Normal distribution.
-   exponential:<lambda>       Exponential distribution.
-   pareto:<scale>,<shape>     Generalized Pareto distribution.
-   gev:<loc>,<scale>,<shape>  Generalized Extreme Value distribution.
+   [fixed:]<value>              Always generates <value>.
+   uniform:<max>                Uniform distribution between 0 and <max>.
+   normal:<mean>,<sd>           Normal distribution.
+   exponential:<lambda>         Exponential distribution.
+   pareto:<loc>,<scale>,<shape> Generalized Pareto distribution.
+   gev:<loc>,<scale>,<shape>    Generalized Extreme Value distribution.
 
-   fb_key                     ETC key-size distribution from [1].
-   fb_value                   ETC value-size distribution from [1].
-   fb_ia                      ETC inter-arrival distribution from [1].
+   To recreate the Facebook \"ETC\" request stream from [1], the
+   following hard-coded distributions are also provided:
+
+   fb_value   = a hard-coded discrete and GPareto PDF of value sizes
+   fb_key     = \"gev:30.7984,8.20449,0.078688\", key-size distribution
+   fb_ia      = \"pareto:0.0,16.0292,0.154971\", inter-arrival time dist.
 
 [1] Berk Atikoglu et al., Workload Analysis of a Large-Scale Key-Value Store,
     SIGMETRICS 2012