ibacm(1)ibacmibacm(1)NAMEibacm - address and route resolution services for InfiniBand.
SYNOPSISibacmDESCRIPTION
The IB ACM implements and provides a framework for name, address, and
route (path) resolution services over InfiniBand. It is intended to
address connection setup scalability issues running MPI applications on
large clusters. The IB ACM provides information needed to establish a
connection, but does not implement the CM protocol.
A primary user of the ibacm service is the librdmacm library. This
enables applications to make use of the ibacm service without code
changes or needing to be aware that the service is in use. librdmacm
versions 1.0.12 - 1.0.15 can invoke IB ACM services when built using
the --with-ib_acm option. Version 1.0.16 and newer of librdmacm will
automatically use the IB ACM if it is installed. The IB ACM services
tie in under the rdma_resolve_addr, rdma_resolve_route, and rdma_getad‐
drinfo routines. For maximum benefit, the rdma_getaddrinfo routine
should be used, however existing applications should still see signifi‐
cant connection scaling benefits using the calls available in librdmacm
1.0.11 and previous releases.
The IB ACM is focused on being scalable and efficient. The current
implementation limits network traffic, SA interactions, and centralized
services. ACM supports multiple resolution protocols in order to han‐
dle different fabric topologies.
The IB ACM package is comprised of two components: the ibacm service
and a test/configuration utility - ib_acme. Both are userspace compo‐
nents and are available for Linux and Windows. Additional details are
given below.
QUICK START GUIDE
1. Prerequisites: libibverbs and libibumad must be installed. The IB
stack should be running with IPoIB configured. These steps assume that
the user has administrative privileges.
2. Install the IB ACM package. This installs ibacm, ib_acme, and
init.d scripts.
3. Run 'ibacm' as administrator to start the ibacm daemon.
4. Optionally, run 'ib_acme -d <dest_ip> -v' to verify that the ibacm
service is running.
5. Install librdmacm, using the build option --with-ib_acm if needed.
The librdmacm will automatically use the ibacm service. On failures,
the librdmacm will fall back to normal resolution.
6. You can use ib_acme -P to gather performance statistics from the
local ibacm daemon to see if the service is working correctly.
NOTES
ib_acme:
The ib_acme program serves a dual role. It acts as a utility to test
ibacm operation and help verify if the ibacm service and selected pro‐
tocol is usable for a given cluster configuration. Additionally, it
automatically generates ibacm configuration files to assist with or
eliminate manual setup.
ibacm configuration files:
The ibacm service relies on two configuration files.
The ibacm_addr.cfg file contains name and address mappings for each IB
<device, port, pkey> endpoint. Although the names in the
ibacm_addr.cfg file can be anything, ib_acme maps the host name and IP
addresses to the IB endpoints. If the address file cannot be found,
the ibacm service will attempt to create one using default values.
The ibacm_opts.cfg file provides a set of configurable options for the
ibacm service, such as timeout, number of retries, logging level, etc.
ib_acme generates the ibacm_opts.cfg file using static information. If
an option file cannot be found, ibacm will use default values.
ibacm:
The ibacm service is responsible for resolving names and addresses to
InfiniBand path information and caching such data. It should execute
with administrative privileges.
The ibacm implements a client interface over TCP sockets, which is
abstracted by the librdmacm library. One or more back-end protocols
are used by the ibacm service to satisfy user requests. Although the
ibacm supports standard SA path record queries on the back-end, it also
supports a resolution protocol based on multicast traffic. The latter
is not usable on all fabric topologies, specifically ones that may not
have reversible paths or fabrics using torus routing. Users should use
the ib_acme utility to verify that multicast protocol is usable before
running other applications.
Conceptually, the ibacm service implements an ARP like protocol and
either uses IB multicast records to construct path record data or
queries the SA directly, depending on the selected route protocol. By
default, the ibacm services uses and caches SA path record queries.
Specifically, all IB endpoints join a number of multicast groups. Mul‐
ticast groups differ based on rates, mtu, sl, etc., and are priori‐
tized. All participating endpoints must be able to communicate on the
lowest priority multicast group. The ibacm assigns one or more
names/addresses to each IB endpoint using the ibacm_addr.cfg file.
Clients provide source and destination names or addresses as input to
the service, and receive as output path record data.
The service maps a client's source name/address to a local IB endpoint.
If a client does not provide a source address, then the ibacm service
will select one based on the destination and local routing tables. If
the destination name/address is not cached locally, it sends a multi‐
cast request out on the lowest priority multicast group on the local
endpoint. The request carries a list of multicast groups that the
sender can use. The recipient of the request selects the highest pri‐
ority multicast group that it can use as well and returns that informa‐
tion directly to the sender. The request data is cached by all end‐
points that receive the multicast request message. The source endpoint
also caches the response and uses the multicast group that was selected
to construct or obtain path record data, which is returned to the
client.
The current implementation of the IB ACM has several additional
restrictions:
- The ibacm is limited in its handling of dynamic changes. ibacm must
be stopped and restarted if a cluster is reconfigured.
- Cached data does not timed out and is only updated if a new resolu‐
tion request is received from a different QPN than a cached request.
- Support for IPv6 has not been verified.
- The number of addresses that can be assigned to a single endpoint is
limited to 4.
- The number of multicast groups that an endpoint can support is lim‐
ited to 2.
SEE ALSOibacm(7)ib_acme(1), rdma_cm(7)ibacm 2012-03-01 ibacm(1)
