RDMA-bench

A framework to understand RDMA performance. This is the source code for our USENIX ATC 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+ with Mellanox OFED 2.4+. Some benchmarks have been tested with CentOS 7 with OpenFabrics OFED.
memcached, libmemcached-dev, libmemcached-tools
libnuma-dev (numactl-devel on CentOS)

Required settings

All benchmarks require one server machine and multiple client machines. Every benchmark is contained in one directory.

The number of client machines required is described in each benchmark's README file. The server will wait for all clients to launch, so the benchmarks won't make progress until the correct number of clients are launched.
Modify HRD_REGISTRY_IP in run-servers.sh and run-machines.sh to the IP address of the server 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

Benchmark description

The benchmarks used in the paper are described below. This repository contains a few other benchmarks as well.

Benchmark	Description
`herd`	An improved implementation of the HERD key-value cache.
`mica`	A simplified implementation of MICA.

`atomics-sequencer`	Sequencer using one-sided fetch-and-add. Also emulates DrTM-KV.
`ws-sequencer`	Sequencer using HERD RPCs (UC WRITE requests, UD SEND responses).
`ss-sequencer`	Sequencer using header-only datagram RPCs (i.e., UD SENDs only).

`rw-tput-sender`	Microbenchmark to measure throughput of outbound READs and WRITEs.
`rw-tput-receiver`	Microbenchmark to measure throughput of inbound READs and WRITEs.
`ud-sender`	Microbenchmark to measure throughput of outbound SENDs.
`ud-receiver`	Microbenchmark to measure throughput of inbound SENDs.

`rw-allsig`	WQE cache misses for outbound READs and WRITEs.

Implementation details

`libhrd`

The libhrd library is used to implement all benchmarks. It consists of convenience functions for initial RDMA setup, such as creating and connecting QPs, and allocating hugepage memory.

Memcached

Distributing QP information (required for connection setup in connected transports, and routing in datagram transports) requires a temporary out-of-band communication channel. To simplify this process, we use a memcached instance to publish (e.g., hrd_publish_conn_qp()) and pull QP information (e.g., hrd_get_published_qp) using global QP names.

Client connection logic

The code was written to work on a cluster that has dual-port NICs, but the switch connectivity does not allow cross-port communication. Using both ports in this constrained environment makes the initial QP connection setup slightly complicated. All benchmarks also work on single-port NICs. Usually, we use the following logic while setting up connections:

There are N client threads in the system and each client thread uses Q QPs.
The server has num_server_ports ports starting from port base_port_index. Similarly, clients have num_client_ports. The base_port_index may be different for server and clients.
On the CIB cluster, port i on a NIC can only communicate with port i on other NICs. So base_port_index must be same for clients and server, and num_client_ports == num_server_ports.
One server thread (the master thread in case there are worker threads) creates N * Q QPs on each server port. For applications requiring a request region, only one memory region is created and registered with all of the num_server_ports control blocks. Only some of these QPs actually get used by clients.
Client threads have a global index clt_i. Each client thread uses a single control block and creates all its QPs on port index (using base base_port_index) clt_i % num_client_ports. It connects all these QPs to QPs on server port indexed clt_i % num_server_ports (using the server's base_port_index). This works for both CIB, and Apt and Intel clusters that support any-to-any communication between ports.

Selective signaling logic

Most benchmarks post one signaled work request per UNSIG_BATCH work requests. This is done to reduce CQE DMAs. With UNSIG_BATCH = 4, a sequence of work requests looks as follows. Note that a work request is not post()ed immediately; it is added to a list and posted when the number of work requests in the list equals postlist.

wr 0 -> signaled
wr 1 -> unsignaled
wr 2 -> unsignaled
wr 3 -> unsignaled
	Poll for wr 0's completion. A postlist should have ended.
wr 4 -> signaled
...
wr 5 -> unsignaled
	Poll for wr 4's completion. Another postlist should have ended.

This imposes 2 requirements:

Postlist check: postlist <= UNSIG_BATCH. We poll for a completion before queueing work request UNSIG_BATCH + 1. If postlist > UNSIG_BATCH, nothing will have been posted at this point, so polling will get stuck.
Queue capacity check: HRD_Q_DEPTH >= 2 * UNSIG_BATCH. With the above scheme, up to 2 * UNSIG_BATCH - 1 work requests can be un-ACKed by the QP. With a QP of size N, N - 1 work requests are allowed to be un-ACKed by the InfiniBand/RoCE specification.

Contact

Anuj Kalia (akalia@cs.cmu.edu)

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.

HKUST-SING/rdma_bench