FaSST

Fast, scalable, and simple transactions over RDMA. For historical reasons, FaSST is called HoTS in this codebase. This is the source code for our USENIX OSDI paper.

Required hardware and software

• InfiniBand HCAs
  • RoCE HCAs have not been tested, but should work with minor modifications. See is_roce() usage in HERD for details.
• Ubuntu 12.04+
• Mellanox OFED 2.4+
• memcached, libmemcached-dev, libmemcached-tools
• libnuma-dev (numactl-devel on CentOS)
• Boost 1.6 (earlier and newer versions not tested)
  • Depends on python-dev and libbz2-dev
• libpapi-dev
• gcc-5.3 (C++11 support required; gcc-4.8 works but can have ~10% lower performance)

Building the benchmarks

There is no top-level Makefile. Each benchmark in the app directory is built separately with the Makefile in the benchmark's directory.

Required settings

All benchmarks run in a symmetric setting with multiple machines (i.e., there are no designated server or client machines. Every benchmark is contained in one directory.

• Benchmark parameters are contained in a json file in the experiment directory.
• Modify HRD_REGISTRY_IP in run-servers.sh to the IP address of server 0's machine. This machine runs a memcached instance that is used as a queue pair registry.
• Allocate hugepages on the NIC's socket at the server machine. On our machines, the NIC is attached to socket 0, so all benchmark scripts bind allocated memory and threads to socket 0. On Ubuntu systems, create 8192 hugepages on socket 0 using:

	sudo echo 8192 > /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages

Code structure

Directory	Description
libhrd	An RDMA management and communication library.
rpc	FaSST RPCs. This should depend only on libhrd and hots.h.
mica2	A C++ implementation of a MICA hash table.
datastore	Interface of FaSST datastores, and RPC handlers for MICA.
tx	The transaction protocol.
logger	The commit record logger for transactions.
mappings	Helper functions to determine primary/backup/log machine numbers.
app	Benchmarks (TATP, SmallBank, RPC microbenchmark, packet loss test).
drivers	Modified libmlx4 and libmlx5 drivers with the cheap RECV posting optimization.
loss\_study	The code and output logs used in our InfiniBand packet loss study.
other	Contains a DCT benchmark.

Instructions to run on Emulab's Apt cluster:

• Set HRD_REGISTRY_IP to node-1's IP address
• Reduce HRD_MAX_INLINE to 60 bytes
• Reduce num_ports to 1
• Change the thread-to-core binding from thread i -> core 2 * i to thread i -> core i

Acknowledgements

Hyeontaek Lim's MICA code is used directly and in a modified way.

Contact

Anuj Kalia ([email protected])

License

	Copyright 2016, Carnegie Mellon University

    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at

        http://www.apache.org/licenses/LICENSE-2.0

    Unless required by applicable law or agreed to in writing, software
    distributed under the License is distributed on an "AS IS" BASIS,
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.