Updated Branches: refs/heads/trunk 5922de147 -> 5e758b849
GIRAPH-622 Project: http://git-wip-us.apache.org/repos/asf/giraph/repo Commit: http://git-wip-us.apache.org/repos/asf/giraph/commit/5e758b84 Tree: http://git-wip-us.apache.org/repos/asf/giraph/tree/5e758b84 Diff: http://git-wip-us.apache.org/repos/asf/giraph/diff/5e758b84 Branch: refs/heads/trunk Commit: 5e758b849e817d846c5a39b57bb07b72eaf40f7d Parents: 5922de1 Author: Claudio Martella <[email protected]> Authored: Wed May 8 01:51:13 2013 +0200 Committer: Claudio Martella <[email protected]> Committed: Wed May 8 01:51:13 2013 +0200 ---------------------------------------------------------------------- src/site/xdoc/ooc.xml | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/giraph/blob/5e758b84/src/site/xdoc/ooc.xml ---------------------------------------------------------------------- diff --git a/src/site/xdoc/ooc.xml b/src/site/xdoc/ooc.xml index 2887a0d..a113d3b 100644 --- a/src/site/xdoc/ooc.xml +++ b/src/site/xdoc/ooc.xml @@ -42,7 +42,7 @@ under the License. <section name="Out-of-core Messages"> <p>When running out-of-core messages, Giraph will keep only a limited number of messages in memory, while the others will be stored to local disk(s). This feature can be enabled with parameter "giraph.useOutOfCoreMessages=true" (disabled by default), while the number of messages is controlled by parameter "giraph.maxMessagesInMemory=N" (with default value 1000000). With this feature, Giraph will keep in memory the incoming messages into an in-memory store. When the store exceeds the chosen number of messages, the content of the store will be spilled to disk, and a new empty in-memory store will be instantiated. This process produces a number of files on disk, depending on the number of messages produced during a superstep. During the vertex computation the files will be read sequentially, and the messages for each vertex will be concatenated and fed to the vertex. Both for reading and writing, files are accessed sequentially. </p> - <p>Also out-of-core messages can take advantage of multiple disks, as parameter "giraph.messagesDirectory" (with default "_bsp/_messages/") can accept a comma-separated list of paths. It is possible to control the buffers used for i/o with parameter "giraph.messagesBufferSize=#Bytes" (with default value 8192). + <p>Also out-of-core messages can take advantage of multiple disks, as parameter "giraph.messagesDirectory" (with default "_bsp/_messages/") can accept a comma-separated list of paths. It is possible to control the buffers used for i/o with parameter "giraph.messagesBufferSize=#Bytes" (with default value 8192). </p> </section> <p>It is difficult to decide a general policy to use out-of-core capabilities, as it depends on the behavior of the algorithm and the input graph. The exact number of partitions and messages to keep in memory depends on the cluster capabilities, the number of messages produced per superstep, and number of active vertices per superstep. Moreover, it depends on the type and size of vertex values and messages. For example, algorithms such as Belief Propagation tend to keep large vertex values, while algorithms such as clique computations tend to send large messages along. Hence, it depends on your algorithm what feature to rely on more.
