[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Assumptions we'll rely on for MDC2
We discussed the various requirements for the Storage Manager
components for MDC2, and want to make explicit the assumptions we
are relying on. This is mainly to verufy that we all agree on
these assumptions.
1) Each event header will contain information about the event
components. Thus, in response to a multi-component request from an
EI, it is sufficient for the Storage Manager to return only the
event_oids list. The user code will get the OIDs for each
component from the event header.
Note: as explained in my last message the response to a request
from an EI has the form:
[file_c(i), file_c(j), ...]: {list of event_oids},
where file_c(i) is the ID of a file for component i that contains
all the event components in the {list of event_oids}. Similarly,
for file_c(j), etc.
We call [file_c(i), file_c(j), ...] a file bundle. There are
multiple file bundles for each multi-component query, where some
files may appear in multiple file bundles.
2) A response to a multi-component request will be made only after
all the component files were cached.
3) For component files that are not stored in Objectivity, there
will be an empty directory in Objectivity. Thus, each
non-objectivity file will have a file_oid, and its names can be
gotten from the objectivity directory, just as for Objectivity
files.
4) Files can be stored in multiple directories. Thus, for each
PFTP the file location (path) will be determined dynamically.
5) The cache (using multiple disks) will be managed by a
multi-disk file manager (such as veratose). Thus, the entire cache
will be viewed as a single cache by the Storage manager.
Comments?
Arie.