Rough Notes by Les Cottrell June 26, 2003
Showed a UML diagram based on CIM, based on the previous day's NMWG presentations and the NMWG schema. There are physical elements and logical elements (for software) to define the static information. Tied into this are classifications that tell whether it is an OS, a computer system, firmware, is it a hub a switch, what is it dedicated to (e.g. to printing). One can also define owners.
She is using diagnostic services to relate to network measurements. Managed elements are what is to be tested. . DiagSetting decides how the measurement is made (e.g. make pings every 30 mins). LogOptions allows you to specify the amount of output, whether to do intensive or just a rough estimate. JobSettingData determines when you run. NetworkPacketTestSetting is where PacketSize, NumberofPaclets, packetSpacing, packetGap, lossThreshold, testCharacteristics would be kept. DiagnosticServiceRecord allows specifying how long to keep the data, how much confidence to put on the result. Can have sub records, e.g. PathBandwidthServiceRecord might be a sub-class of DiagnosticServiceRecord, further PathAchievableBandwidth could be a further subclass of PathBandwidthServiceRecord to allow defining number of streams, TCP|UDP ..., disk etc. For PathDelayServiceRecord has a roundTrip boolean, percentiles, min, avg, max, stdev.
CIM wants to map this out to XML schema. But first wants the Grid schema to solidify.
Typical queries are for a specified source:target, time frame, and measurement type.
Keith Jackson is concerned about including the definitions of computer, OS' etc. which is defined elsewhere in the grid.
To him network measurements are another form of meta data, and OGSI provides standard mechanisms for accessing meta-data. OGSI has a service data element that is just a wrapper around a service specific meta-data and state, and ServiceDataValue element can contain arbitary XML. Then can make intelligent queries.
This alows usage of standard OGSI service tooling: service lifecycle, security, notification and creation services. Can use standard query engines such as XPATH or Xquery to find by element name. Can have multiple bindings to the query function. This allows returning of SDE's (Service Data Elements) in multiple encodings, e.g. an efficient binary format.
Efficiency depends on depth of schemas and complexity of the query. Today it is probably much less efficient than SQL. The technology is coming on. Xpath is not as rich as SQL, if need detailed queries then may need Xquery. May need filters added as shims to provide more efficiency (both for input and output).
It may be useful to collect use cases for queries that people are making. Dave Martin of IBM and Martin Stouffer of LBL volunteered to lead this effort.
We discussed some of the open issues in the schema (in red in http://www-didc.lbl.gov/NMWG/NMWG-profile.html).
The timeAccuracy was agreed to mean the estimated difference between the timestamp and true time (e.g. a perfectly accurate GPS clock).
There was an inconclusive discussion on hoplist and how to represent traceroutes.