I shall be going to SC2012 next month, I plan on hitting a few of the storage vendors for possible collaborations and flagging to them that we’re on the look out for storage systems. One of the first observation that the reader will note is “where is that link between HPC and Digital Preservation and Archiving”. It’s probably not obvious to most people, one of the big problems in the area of preservation and archiving is the the amount of data involved and the varied types of data. This is not taking into account of the issues with data access patterns.
Given that a preservation and archiving project will want to provide a trusted system, the system will want to read out every single byte that was put in to verify that the data is correct at somepoint (usually with some form of hashing).
Reading data out and checking that it’s correct serially probably isn’t the smartest solution. Nor is copying the data into 2-3 locations (where each site is maintaining 2-3 copies for backups and redundancy). The current and seemingly most popular solutions is to dump the data to a few offsite locations (such as S3 or SWIFT) compatible storage systems, then just hoping for the best that if anyone of the sites is down or corrupted there site can be restored from the other sites or from a backup. I need to delve deeper into the storage and data-distribution strategies that some of the bigger projects are taking. There has to be a smarter way of storing and preserving data without having to make copies of things.
I’ve often wondered how projects manage to copy/move data across storage providers in a reasonable amount of time without needing to wheel a few racks of disks around. It would also be interesting to see the error rates of these systems and how often errors are corrected. If they are corrected what is the computational cost of doing this.
If you have a multi-terabyte archive the problem isn’t too bad, the more typical case these days might be in the order of the low hundreds of terabytes. I could only imagine what lager scale sites must deal with. I’m still not a fan of moving a problem from a local site to a remote site as it often shows that there is a lack of understanding to the problem. Storage in the preservation and archiving domain will probably turn into an IO and compute intensive operation at some point, especially if you want to do something with the data.
Share!
See Also
- slurm bank that big script for banking in slurm
- cports for building applications and libraries for hpc systems
- a poor mans nas device with ceph
- scientific linux 6 build environment for ceph
- installing ceph on sl6