Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Migrated to Confluence 5.3

...

The coordination layer offers communication through a message broker and a data transmission via https. This is illustrated in the below figure. In general communication is initiated and actions are coordinated via the message broker, while the actual transmission of data is done via data transmission.

There are is no intelligence in the Coordination Layer and it is the clients that are responsible for choices that involves intelligence e.g. in chooing choosing an alternative pillar for request of data (according to Service Level Agreement).

...

All communication via the Coordination layer starts via a Topic that is send to all. It is up to the individual subscribers to determine whether it is a message that concerns them, for example a Get GetTime message does only concern the pillars involved in a certain service level agreement.

<TBD> An example can be found in http://wiki.statsbiblioteket.dk/BitmagasinWiki/Sekvens1

DIAGRAM NOT UPDATED!

Atomic messaging in simple operations

...

An operation is defined as the communication that takes place in order to make an operation on the bit repository. Examples are get and put of data.

Operations must be as atomic and simple as possible. this They will be based on simple message primitives and data transmission. A primitive is a single request message to the system and the following replies. The primitives are defined on basis of atomic weldefined messages. This serves to give simplicity, and to get as little overlap between different operations as possible.

Using Get as example

The Get operation can begin by asking pillars who will be the fastes, or it can start by actually sending a get message for a specific default pillar.

In the strict interpretation of single operations this will be seen as two operations

  1. GetTime
  2. Get

An the difference between the two get scenarios will be that teh first will both consist of a GetTime and a Get opertation, while the later will only consist of a Get.

The below figure illustrates the messages of the two operations illustrated, where messages belonging to the same communication in an operation is placed edge to edge. note that it only includes messages on the message bus, i.e. excluding the actual data transmission.

Image Removed

Since each communication is started via a topic this means that the GetTime and the Get is sent to a Topic. This is illustrated in the below sequence diagram

<TBD> See http://wiki.statsbiblioteket.dk/BitmagasinWiki/Sekvens1

The advantages are:

  • the Get operation works similar for the two scenarios
  • it will be easy to handle shift in services, e.g. exchanging one primary pillar with another, exchaging pillars in a SLA etc.

The disadvantages can be:

...

An example of the Get operation is given in the below figure.

Image Added

Here the full communication will result in an operation. The message communication is divided into two parts, each representing a primitive (identification of pillars and the actual get).

Note that the primitive for identification of pillars can cover request to several pillars, while the primitive for the actual get part only concerns one pillar.

Note also that there can be a number of different reponses, in cases where the pillar want to give information of the progress of the ioperation at pillar level. This can for instance be interesting for DVD pillars where there are several possibly time consuming phases in finding the data - even more so for a put operation.

Conversation Id

We generate a unique conversation id at initiation of a conversation such that it is possible to identify which conversation a message belongs to.

Reuse and overlap between operations

The operations are design to have as many similarities and overlapping definitions as possible. For instance the getChecksums must work simmilar to Get. Paging of data and parameters must work in the same way. See dicussion on http://kb-dk.atlassian.net/wiki/display/BITMAG/Coordination+layer+design+discussion#Coordinationlayerdesigndiscussion-Overlapmedprotokolspecifikation.

Paging af requests

It must be possible to make paging of request parameters. Paging of parameters can for instance be the case for object ids in a random check of files. It is needed since we cannot assume that clients can make prior pagination. See dicussion on http://kb-dk.atlassian.net/wiki/display/BITMAG/Coordination+layer+design+discussion#Coordinationlayerdesigndiscussion-Pagingafrequests.

Software

Independence

In the design we need to ensure that we are as independent of the software chosen as possible. If special features are used, e.g. for optimisation, then the design must be flexible enough to skip use of the special features at a later stage.

...

  • it is sufficient to identify a data transmission transaction by a token
  • the token is sufficient information as basis for completion of a data transmission

...

Tokens are used to til designation of where the data can be put to or downloaded from.

Issuing of token are alsways done by client in client/pillar communications. Creation of the actual connection is done by the pillar.

Paging

Paging is in the first versions manage managed by a simple mecanism mechanism where a segment of a file is identified by parameters. Any Get command can contain an offset along with a length parameter, which defines the segment of the file. this is also sufficient to handle receival of divided files, and thus enabling rudimentary support of restart of an interrupted transmission.

...