diff --git a/figures/Makefile b/figures/Makefile
index 74e0cca..991b066 100644
--- a/figures/Makefile
+++ b/figures/Makefile
@@ -41,7 +41,7 @@ CSVS        := $(addprefix $(DERIVED)/, runs.csv rtts.csv idts.csv firstflow_bin
 PDFS        := $(addprefix $(OUT)/, $(addsuffix .pdf, $(FIG_STEMS)))
 TEXS        := $(addprefix $(OUT)/, $(addsuffix .tex, $(FIG_STEMS)))
 
-.PHONY: all derive figures tikz sanity values pdfs tex clean datacenter-fq datacenter-fq-tikz direct-link direct-link-tikz slowstart slowstart-tikz bandwidths bandwidths-tikz all-experiments all-experiments-tikz
+.PHONY: all derive figures tikz sanity values pdfs tex clean datacenter-fq datacenter-fq-tikz direct-link direct-link-tikz slowstart slowstart-tikz bandwidths bandwidths-tikz flows flows-tikz all-experiments all-experiments-tikz
 .DELETE_ON_ERROR:
 
 all: all-experiments
@@ -185,8 +185,18 @@ bandwidths:
 bandwidths-tikz:
 	$(MAKE) tex $(BANDWIDTHS_ARGS)
 
-all-experiments: datacenter-fq direct-link slowstart
-all-experiments-tikz: datacenter-fq-tikz direct-link-tikz slowstart-tikz
+FLOWS_ARGS = EXP=flows SETUP=flows \
+	  SOLUTIONS="2:2 4:4 8:8 16:16 30:30" \
+	  FIG_STEMS="flows_idt_cdf"
+
+flows:
+	$(MAKE) pdfs $(FLOWS_ARGS)
+
+flows-tikz:
+	$(MAKE) tex $(FLOWS_ARGS)
+
+all-experiments: datacenter-fq direct-link slowstart bandwidths flows
+all-experiments-tikz: datacenter-fq-tikz direct-link-tikz slowstart-tikz bandwidths-tikz flows
 
 # Slowstart: standalone, reads raw text files directly (no aggregator).
 SLOWSTART_OUT := ../out/slowstart
diff --git a/figures/fig_flows_idt_cdf.R b/figures/fig_flows_idt_cdf.R
new file mode 100644
index 0000000..f1a06e4
--- /dev/null
+++ b/figures/fig_flows_idt_cdf.R
@@ -0,0 +1,32 @@
+#!/usr/bin/env Rscript
+source("common.R")
+
+FLOW_LEVELS <- c("2", "4", "8", "16", "30")
+
+parser <- fig_parser(description = "Per-flow IDT CDF across flow counts (TSO Pacing)")
+parser$add_argument("--sample", type = "integer", default = 20000,
+                    help = "downsample to N points per flow count (0 = none)")
+parser$add_argument("--seed", type = "integer", default = 1)
+args <- parser$parse_args()
+
+set.seed(args$seed)
+idts <- read_csv(file.path(args$data, "idts.csv"), show_col_types = FALSE) %>%
+  mutate(nflows = factor(as.character(solution), levels = FLOW_LEVELS))
+if (args$sample > 0) {
+  idts <- idts %>% group_by(nflows) %>% slice_sample(n = args$sample) %>% ungroup()
+}
+
+p <- ggplot(idts, aes(x = idt_us, colour = nflows))
+p <- p + stat_ecdf(linewidth = 0.9, pad = FALSE, key_glyph = "rect")
+p <- p + scale_x_log10(labels = label_comma())
+p <- p + scale_y_continuous(labels = label_pct())
+p <- p + scale_colour_viridis_d(end = 0.9, name = "Flows")
+p <- p + labs(x = paste0("Inter-departure time within flow (", label_us(), ")"), y = "CDF")
+p <- p + theme_paper()
+p <- p + theme(legend.text = element_text(size = rel(0.8)), legend.key.size = unit(6, "pt"))
+
+if (args$zoom) {
+  p <- p + coord_cartesian(xlim = c(3, 1000), ylim = c(0.9, 1.0))
+}
+
+save_figure(p, args)