add brief readme

examples/composition/README.txt

+The examples in this directory benchmark the overhead of relaying to RPCs to co-located
+services in various configurations.
+
+<NOTE: ignore performance results in these examples; they were taken from a debugging build>
+
+The following example will compare the latency of issuing RPCs (with 8 MiB data transfers)
+directly to a service vs the latency of the same RPC relayed through a delegator service.
+The two daemon instances do not need to run on the same node, but you need to use another
+transport like CCI, BMI, or OFI to relay between remote nodes.
+
+==========================
+
+$ ./composed-svc-daemon na+sm:// data-xfer
+# accepting RPCs on address "na+sm://24151/0"
+
+$ ./composed-svc-daemon na+sm:// delegator
+# accepting RPCs on address "na+sm://24180/0"
+
+$ ./composed-benchmark na+sm://24180/0 na+sm://24151/0 100
+# DBG: starting data_xfer_read() benchmark.
+# DBG:    ... DONE.
+# <op> <min> <avg> <max>
+direct  0.001103    0.001333    0.005824
+# DBG: starting composed_read() benchmark.
+# DBG:    ... DONE.
+# <op> <min> <avg> <max>
+composed    0.001103    0.001751    0.005824
+# DBG:    ... DONE.
+Shutting down delegator server.
+Shutting down data_xfer server.
+
+==========================
+
+Alternatively, you can run the two services in the same executable, in which case the relay
+will bypass the network layer of Mercury entirely.
+
+==========================
+
+$ ./composed-svc-daemon na+sm:// data-xfer,delegator
+# accepting RPCs on address "na+sm://24259/0"
+
+$ ./composed-benchmark na+sm://24259/0 na+sm://24259/0 100
+# DBG: starting data_xfer_read() benchmark.
+# DBG:    ... DONE.
+# <op> <min> <avg> <max>
+direct  0.001215    0.001514    0.010304
+# DBG: starting composed_read() benchmark.
+# DBG:    ... DONE.
+# <op> <min> <avg> <max>
+composed    0.001215    0.001756    0.010304
+# DBG:    ... DONE.
+Shutting down delegator server.
+
+