# HG changeset patch
# User shade
# Date 1494580303 -7200
# Node ID b653734241e897b824cdfd47f79fa54937055244
# Parent  30a5ac8fea97cddd07378b1c0aa8dec280c660a5
Add very basic README.

diff -r 30a5ac8fea97 -r b653734241e8 README.md
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/README.md	Fri May 12 11:11:43 2017 +0200
@@ -0,0 +1,26 @@
+Garbage Collection Benchmarks
+------------------------------------------------------------------------------------
+
+This project collects the interesting workloads for Garbage Collection performance
+measurement. The workloads are exercising the specific modes where GC may find
+themselves. Note that most workloads here are inherently synthetic, and they showcase
+the GC performance in several modes, not necessarily present in all applications
+and all workloads.
+
+The suite is organized as to stress these behaviors:
+
+ A. The dependency of object lifecycle on GC performance. Different applications
+    exhibit different object lifetimes. At one extreme, there are heavily generational
+    workloads where the most objects die young. At another extreme, there are
+    workloads where almost no young objects die, and instead the oldest objects
+    continuously die (think LRU caches).
+
+ B. Allocation pressure. There are several distinct modes for GC to operate in:
+      1. No allocation. In this case, the workloads measure the runtime overheads
+         of mutator-GC interface.
+      2. Extreme allocation. The adversarial workloads that do not do any work,
+         but only allocate objects. This measures how GC is able to recover under
+         significant pressure.
+      3. Normal allocation (where "normal" is defined as a few points in-between
+         no and extreme allocation)
+