Skip to content

Commit

Permalink
manpages: LDMS description
Browse files Browse the repository at this point in the history
  • Loading branch information
oceandlr committed Sep 15, 2014
1 parent ad19aaa commit 913a2e1
Showing 1 changed file with 32 additions and 30 deletions.
62 changes: 32 additions & 30 deletions util/man/LDMS_QuickStart
Original file line number Diff line number Diff line change
Expand Up @@ -7,51 +7,53 @@ LDMS_QuickStart - man page for Quick Start of LDMS

.SH INTRODUCTION
LDMS is the Lightweight Distributed Metric Service. LDMS is a distributed data collection, transport, and storage tool that supports a wide variety of configuration options.
The three major functional components are described below.
The host daemon is the same base code in all cases; differ-
entiation is based on configuration of plugins for sampling or
storage and on configuring aggregation of data from other host daemons.
There are three main functional components described below.
.PP
.I
Samplers
run one or more sampling plugins that periodically sample data of interest on monitored nodes. The sampling
frequency is user defined and can be changed on the fly.
Sampling plugins are written in C. Each plugin defines a
collection of metrics called a metric set. Multiple plugins can
be simultaneously active. By default each sampling plugin
operates independently and asynchronously with respect to
all others. Memory allocated for a particular metric set is
overwritten by each successive sampling and no sample history
is retained within a plugin or the host daemon.
run one or more plugins that periodically sample data of interest.
Each plugin defines a group of metrics called a metric set.
The sampling frequency is user defined and can be dynamically changed.
A host can simultaneously run multiple plugins.
Configuration flags determine whether the sampling plugins run synchronously or asynchonously
(both on a host and across hosts). Memory allocated for a particular metric set is overwritten by each
successive sampling. The host daemon does not retain sample history;
plugins do not typically retain history, but can be written to do so.
.PP
.I
Aggregators
Aggregators
collect data in a pull fashion from samplers
and/or other aggregators. As with the sampler, the frequency of
collection is user defined and operates independently of other
and/or other aggregators. The collection frequency
is user defined and operates independently of other
collection operations and sampling operations. Distinct metric
sets can be collected and aggregated at different frequencies.
Unlike the samplers, the aggregation schedule cannot be altered once set without restarting the aggregator. The number of
hosts collected from by a single aggregator is referred to as the
fan-in. The maximum fan-in varies by transport but is roughly
9,000:1 for the socket transport in general and for the RDMA
transport over Infiniband. It is > 15, 000 : 1 for RDMA over
the Cray Gemini transport. Daisy chaining is not limited to two
levels and multiple aggregators may aggregate from the same
sampler or aggregator ldmsd. Fan-in at higher levels is limited
sets can be collected at different frequencies. Once started, the aggregation schedule cannot
be altered without restarting the aggregator. Fan-in refers to
the number of hosts collected from by a single aggregator.
Maximum fan-in varies by transport but is roughly
9,000:1 for the socket transport and for the RDMA
transport over Infiniband. It is > 15000:1 for RDMA over
the Cray Gemini transport. Daisy chaining is not limited to two levels;
multiple aggregators may aggregate from the same sampler or aggregator ldmsd.
Fan-in at higher levels is limited
by the aggregator host capabilities (CPU, memory, network
bandwidth, and storage bandwidth).
.PP
.I
Storage plugins
write in a variety of formats.
Storage
plugins write in a variety of formats.
Comma Separated Value (CSV) file storage of metric sets
plugins are provided. The
frequency of storage is dependent on the frequency with which
plugins are provided. Storage occurs when a
valid updated metric set data is collected by an aggregator that
has been configured to write that data to storage. Collection of
a metric set whose data has not been updated or is incomplete
does not result in a write to storage in any format.
does not result in a write to storage in any format. The store_derived_csv plugin
can be configured to retain the immediately previous state history and thus store
rate values.

.PP
The host daemon is the same base code in all cases; differentiation is based on configuration of plugins for sampling or
storage and on configuring aggregation of data from other host daemons.


.SH DESCRIPTION
Quick Start instructions for LDMS (Lightweight Distributed Metric Service).
Expand Down

0 comments on commit 913a2e1

Please sign in to comment.