It’s currently 12:18 AM PST, and we’re sitting at 42 minutes away from shutting down the grid for the big asset conversion. I will be posting regular updates on the OSGrid twitter feed. Today we’ll be doing two important upgrades to OSGrid infrastructure – the first will be moving to new hardware for the asset server, the second will be converting to the new FragstoreFS backend.

The first step will be shutting down the grid so we can export the database onto the new server – based on previous exports, this is estimated to take approximately 2 hours. The next step will be importing the database and adjusting the schema for the conversion – this involves changing the indexes to add a numeric incremental index which allows us to resume conversion from specified points. The import will take around one to two hours, and the index conversion will take a similar period of time.

The next step will be firing up the fragstore converter – this is a specialised piece of software we wrote for this, which will reformat the data into the new CAS-aware schema, and move the BLOB records into the filesystem. Conversion speed varies greatly based on the size of the individual assets being processed, but we’re hoping this will take between 9 and 15 hours to complete – we preconverted a large portion of the database already; but the exact speedup we will get from the preconversion is unknown.

Total estimated downtime is 21 hours – but this will hopefully be the last time we need to shut down the grid for such an extended period of time, as future asset moves will be able to be done transparently to grid users (Murphy willing.).