Disterl configurations for experiments.

.gitignore

@@ -11,3 +11,4 @@ results.csv
 rebar3.crashdump
 results-*
 dets/
+*.csv

README.md

@@ -0,0 +1,32 @@
+# ATC'19 Results for Partisan
+
+Here you can find the test harness used for generating all of our results.
+Obviously, your results will differ from ours if you do not use the same
+experimental configuration as we do.
+
+Preliminary results are available [here.](http://partisan.cloud)
+
+This repository includes a microbenchmark suite for Partisan itself, and a
+port of Riak Core (from Basho) ported to Partisan for running experiments
+with a simulated key-value store using three vnodes, and a echo service using
+a single vnode.
+
+Our port of Riak Core is available [here](https://github.com/lasp-lang/riak_core/tree/partisan-support-r21) with a compatibility library available [here](https://github.com/lasp-lang/riak_core_partisan_utils).
+
+## Data Collection Targets
+
+* `bin/microbenchmarks.sh`: Run the microbenchmark suite, Figure 1.
+
+* `bin/microbenchmarks-queueoverhead.sh`: Run the microbenchmark suite for identifying queue overhead introduced by additional connections/queues, Figure 2.
+
+* `bin/microbenchmarks-highlatency.sh`: Run the high-latency microbenchmark suite, Figure 3.
+
+* `bin/microbenchmarks-largepayload.sh`: Run the large-payload microbenchmark suite, Figure 4.
+
+* `bin/throughput-echo.sh`: Run the Echo workload, Figures 5 & 6.
+
+* `bin/throughput-kvs.sh`: Run the KVS workload, Figures 7 & 8.
+
+## Plots
+
+Available in the ```Rscripts``` folder are a number of scripts used to generate the plots included in our final paper.

Rscripts/microbenchmarks-highlatency.R

@@ -0,0 +1,66 @@
+library(ggplot2)
+library(tidyverse)
+library(scales)
+library(shades)
+library(socviz)
+
+# df <- read.csv("C:\\Users\\cmeiklej\\source\\repos\\unir-2019-03-28\\microbenchmarks.csv", header = FALSE)
+# df <- read.csv("C:\\Users\\cmeiklej\\OneDrive\\Desktop\\usenix-2019-camera-ready-results\\microbenchmarks.csv", header = FALSE)
+# df <- read.csv("C:\\Users\\chris\\Documents\\GitHub\\unir\\results-3.csv", header = FALSE)
+df <- read.csv("../microbenchmarks-highlatency.csv", header = FALSE)
+colnames(df) <- c("App", "Transport", "Concurrency", "Channels", "Monotonic", "Parallelism", "Affinity", "Size", "Num", "Latency", "Time")
+df <- transform(df, Time = round(as.numeric(Time) / 1000, digits = 2))
+
+df1 <- subset(df, (
+  # disterl 1 1 false 
+  (Transport == "disterl" & Channels == 1 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 1 1 false
+    (Transport == "partisan" & Channels == 1 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 1 16 false
+    (Transport == "partisan" & Channels == 1 & Parallelism == 16 & Affinity == "false") |
+
+    # partisan 1 16 true
+    (Transport == "partisan" & Channels == 1 & Parallelism == 16 & Affinity == "true") |
+
+    # partisan 4 1 false
+    (Transport == "partisan" & Channels == 4 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 4 16 true 
+    (Transport == "partisan" & Channels == 4 & Parallelism == 16 & Affinity == "true") 
+)
+& (Size == 524288 & Latency == "20" & is.element(Concurrency, c(16, 32, 64, 128))))
+df1$Concurrency <- factor(df1$Concurrency)
+df1$Experiment <- paste("",
+                        ifelse(df1$Affinity == "true" & df1$Transport == "partisan" & df1$Channels == 4 & df1$Parallelism == 16, "partisan parallel affinitized (N = 16)",
+                               ifelse(df1$Affinity == "true" & df1$Transport == "partisan" & df1$Channels == 1 & df1$Parallelism == 16, "partisan parallel affinitized (N = 16)",
+                                      ifelse(df1$Affinity == "false" & df1$Channels == 1 & df1$Transport == "partisan" & df1$Parallelism == 1, "partisan (N = 1)",
+                                             ifelse(df1$Affinity == "false" & df1$Channels == 4 & df1$Transport == "partisan" & df1$Parallelism == 1, "partisan w/ channels (N = 1)",
+                                                    ifelse(df1$Affinity == "false" & df1$Transport == "partisan" & df1$Channels == 1 & df1$Parallelism == 16, "partisan parallel (N = 16)",
+                                                           "disterl"))))))
+
+
+ggplot(data = df1, aes(x = Concurrency, y = Time, fill = Experiment)) +
+  geom_jitter(alpha=0.1, position = position_jitterdodge(jitter.width = 0.1), aes(color = Experiment)) +
+  geom_boxplot(outlier.shape = NA,  notch = TRUE) +
+  # geom_boxplot() + 
+  # theme_grey(base_size = 14) + 
+  ylab("Milliseconds / RT Request") + 
+  xlab("# of Actors") +
+  #scale_y_continuous(trans='log2') + 
+  #annotation_logticks(base = 2, sides = "l") +
+  ggtitle("512KB Payload, 20ms RTT Latency") +
+  theme_classic() +
+  theme(legend.position = c(0.27, 0.86), 
+        panel.grid.major.y = element_line(size = 0.25, linetype = 'solid', colour = "grey60"),
+        panel.grid.minor.y = element_line(size = 0.1, linetype = 'solid', colour = "grey80"),
+        # legend.background = element_rect(color = "black", fill = "grey90", size = 1, linetype = "solid"),
+        plot.title=element_text(size=18),
+        axis.text=element_text(size=16),
+        axis.title=element_text(size=18),
+        legend.title=element_text(size=16, face="bold"), 
+        legend.text=element_text(size=16)
+  )
+
+ggsave("../Microbenchmarks-HighLatency-Final-linear.png", dpi = 400)

Rscripts/microbenchmarks-largepayload.R

@@ -0,0 +1,66 @@
+library(ggplot2)
+library(tidyverse)
+library(scales)
+library(shades)
+library(socviz)
+
+# df <- read.csv("C:\\Users\\cmeiklej\\source\\repos\\unir-2019-03-28\\microbenchmarks.csv", header = FALSE)
+# df <- read.csv("C:\\Users\\cmeiklej\\OneDrive\\Desktop\\usenix-2019-camera-ready-results\\microbenchmarks.csv", header = FALSE)
+# df <- read.csv("C:\\Users\\chris\\Documents\\GitHub\\unir\\results-3.csv", header = FALSE)
+df <- read.csv("../microbenchmarks-largepayload.csv", header = FALSE)
+colnames(df) <- c("App", "Transport", "Concurrency", "Channels", "Monotonic", "Parallelism", "Affinity", "Size", "Num", "Latency", "Time")
+df <- transform(df, Time = round(as.numeric(Time) / 1000, digits = 2))
+
+df1 <- subset(df, (
+  # disterl 1 1 false 
+  (Transport == "disterl" & Channels == 1 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 1 1 false
+    (Transport == "partisan" & Channels == 1 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 1 16 false
+    (Transport == "partisan" & Channels == 1 & Parallelism == 16 & Affinity == "false") |
+
+    # partisan 1 16 true
+    (Transport == "partisan" & Channels == 1 & Parallelism == 16 & Affinity == "true") |
+
+    # partisan 4 1 false
+    (Transport == "partisan" & Channels == 4 & Parallelism == 1 & Affinity == "false") |
+
+    # partisan 4 16 true 
+    (Transport == "partisan" & Channels == 4 & Parallelism == 16 & Affinity == "true") 
+)
+& (Size == 8388608 & Latency == "1" & is.element(Concurrency, c(16, 32, 64, 128))))
+df1$Concurrency <- factor(df1$Concurrency)
+df1$Experiment <- paste("",
+                        ifelse(df1$Affinity == "true" & df1$Transport == "partisan" & df1$Channels == 4 & df1$Parallelism == 16, "partisan parallel affinitized (N = 16)",
+                               ifelse(df1$Affinity == "true" & df1$Transport == "partisan" & df1$Channels == 1 & df1$Parallelism == 16, "partisan parallel affinitized (N = 16)",
+                                      ifelse(df1$Affinity == "false" & df1$Channels == 1 & df1$Transport == "partisan" & df1$Parallelism == 1, "partisan (N = 1)",
+                                             ifelse(df1$Affinity == "false" & df1$Channels == 4 & df1$Transport == "partisan" & df1$Parallelism == 1, "partisan w/ channels (N = 1)",
+                                                    ifelse(df1$Affinity == "false" & df1$Transport == "partisan" & df1$Channels == 1 & df1$Parallelism == 16, "partisan parallel (N = 16)",
+                                                           "disterl"))))))
+
+ggplot(data = df1, aes(x = Concurrency, y = Time, fill = Experiment)) +
+  geom_jitter(alpha=0.1, position = position_jitterdodge(jitter.width = 0.1), aes(color = Experiment)) +
+  geom_boxplot(outlier.shape = NA,  notch = TRUE) +
+  # geom_boxplot() + 
+  # theme_grey(base_size = 14) + 
+  ylab("Milliseconds / RT Request") + 
+  xlab("# of Actors") +
+  #scale_y_continuous(trans='log2') + 
+  #annotation_logticks(base = 2, sides = "l") +
+  ggtitle("8MB Payload, 1ms RTT Latency") +
+  theme_classic() +
+  theme(legend.position = c(0.27, 0.86), 
+        panel.grid.major.y = element_line(size = 0.25, linetype = 'solid', colour = "grey60"),
+        panel.grid.minor.y = element_line(size = 0.1, linetype = 'solid', colour = "grey80"),
+        # legend.background = element_rect(color = "black", fill = "grey90", size = 1, linetype = "solid"),
+        plot.title=element_text(size=18),
+        axis.text=element_text(size=16),
+        axis.title=element_text(size=18),
+        legend.title=element_text(size=16, face="bold"), 
+        legend.text=element_text(size=16)
+  )
+
+
+ggsave("../Microbenchmarks-LargePayload-Final-linear.png", dpi = 400)