User login

Weekly Report -- 24/05/2013




Finished adding simple time series graphs for our switch interface byte count data. Got Brendon's event rendering working with these new graphs too, so we can now see and explore the events detected using the Plunge and ArimaShewhart detectors. They seem to be working reasonably well so far.

The next task I started on was fixing the URLs for the amp-web graphs -- the current setup is graph/// which is not sustainable going forward. Firstly, the metric needs to come first so that we can handle time series that are defined by more than just a source and target, e.g. a direction or an application protocol. Next, instead of explicitly listing the source, target or whatever else describes the time series data, we want to use the unique stream id from within NNTSC. This also avoids the problem of our URLs being really long or containing spaces. Unfortunately, much of the original code was written with only source and target in mind so there's a lot to change to be able to support LPI data, for example.

Developed a new version of libwandevent. There are two main changes in the new version. Firstly, the allocation and management of event structures is all handled internally by libwandevent -- no more filling in event structures and passing them off to libwandevent. The main reason for this is to try and minimise the chance of bugs where the programmer inadvertantly overwrites an existing event, much like the BSOD bug I had last week. However, it does break the existing API so there may be a slightly messy transition period. Secondly, I've added support for epoll so that will now be used instead of select, if available. Switched BSOD server over to use the new libwandevent and it seems to work pretty well.