My Way: CERN
The data generated by the CERN Large Hadron Collider project will exceed every other scientific experiment in history. E&T speaks to the project 'data management czar' Dr Charles Curran.
E&T: Why is it important for CERN to update its data storage facilities?
Charles Curran: We are already probably one of the biggest data holders in the world, with eight petabytes of physics data held on 30,000 cartridges. But because of the LHC project, the number of tape cartridges we need will increase markedly. We are expecting to add in the region of 15,000 to 20,000 tape cartridges every year.
E&T: But presumably it's not just about capturing data?
CC: We have to make sure the data is accessible and keep it organised, so we also have to make sure it is transferable to the next generation of tape media. But checking the quality of old tapes and verifying the integrity of the data they hold has historically made it difficult to keep up with the pace of migration.
E&T: The primary storage medium will be tape, in IBM 3584 and Sun StorageTek StreamLine tape libraries. How will the Tapewise tape management software help CERN solve its storage challenges?
CC: It seems to me that this software package is in many ways a distillation of years of experience of professionals working in the area of tape storage in IT. It can also take nerve to tell a manufacturer that you have suspicions that their products are showing a problem; this tool can provide a 'second opinion' when used to look at possibly problematic pieces of media or drives. That carries some weight, as I believe some of the suppliers use it themselves.
E&T: Anything else?
CC: At the same time, testing a selection of media for its ability to be totally writable when a batch is purchased is often neglected due to pressure to 'get on with it', but Tapewise offers an automated route to do this. Equally, we may well seem to be storing data perfectly well, but a programme to selectively sample the park of tapes that have been written is very important and the package also addresses this in an automated manner.
E&T: How exactly did you perform scanning processes before the Tapewise package?
CC: We used to use an external certification service to sample newly delivered media, but that became impossible as our equipment evolved more quickly than that the service provider had available. Next we used a home-written 'tape scanner' to read complete volumes, and report as far as possible on any difficulties encountered: indeed, we still do this. However, although it is reasonably effective at finding a problem on a tape, it is much less obvious what the very low-level details of the problem are unless you start to try to change the device driver yourself. This is a rather specialist task, not to be undertaken lightly. The Tapewise product reports all details that are available to it.
E&T: How big can the problem of tape integrity be?
CC: For some 10 years now we have to a large extent just crossed our collective fingers, waiting for problems to reveal themselves as we migrated our total data collection - now approaching 10 petabytes - every few years to new drives offering higher data density on the existing media, or to both new drives and new media. These problems may not be very frequent, but we find that approximately 2 per cent of our data volumes do present a problem at the first attempt to migrate the data.
E&T: What does 2 per cent represent in terms of the CERN requirement?
CC: In a collection of around 40,000 tapes to be treated, as we have at present, this is quite a headache when each of these cases will need to be looked at carefully to decide on a recovery strategy.
E&T: There are plenty of those around.
CC: Bear in mind, we are not quite typical of most IT environments: we rarely have two copies of a data file, as the cost is just too great for the volumes we manage and physics data can tolerate losses. Nevertheless, data loss is not pleasant.
E&T: Do you have an ultimate vision or goal you are working towards?
CC: What I would like to see is an environment in which normal IT staff - and not dedicated experts - can ensure the reliable use of our very large collection of physics data for the foreseeable future.
E&T: Does Grid computing have any bearing on your strategy here?
CC: We are expecting that the global use of Grid techniques for data distribution will for the first time provide in principle a second copy of our data somewhere in the world, which will also help enormously (if our collaborators have sufficient budgets!). However, I would want our set of copies to be the most reliably accessible of all and the least costly to maintain.
E&T: Do you have any out-of-work interests that you think help you bring a different perspective to your day job?
CC: With some other colleagues at CERN, I started to learn to play the bagpipes about 25 years ago. Playing this instrument is a wonderful aid to developing a thick skin and a brass neck, which comes in very useful in working in the tape storage area.
E&T: What about the perks of life in snowy Switzerland?
CC: Naturally, in the Geneva area where CERN is sited most people ski and I'm no exception. During the local season, I can even ski briefly at lunchtime and that's a wonderful stress reliever and diversion from the daily problems - it literally cools you down!