The Small Business Operational View from Technology Strategists

Technology Strategists considers backup to be a serious issue -- a defining activity for the Information Technology function as a whole.  Experience of ourselves and others suggest that to a great degree, the seriousness with which the intellectual and data assets of a corporation are protected reflects the care and attention to detail that is shown in other areas -- possibly even a predictor of the organizations success as a whole. It certainly displays the appetite for risk on the part of management.

Technology Strategists is a small business, and however interesting, has few resources to invest in backup operations and technology.  The falling costs of disk drives and need for capacity to accommodate ever more bloated applications, increasing volumes of retained data -- documents and ever bulkier media files has lead to an environment that cannot be effectively backed up.  Unlike disk drives, the cost of removable media and supporting technology has fallen little over the years. And spending many thousands of dollars on a high capacity tape drive and even more on the media (still $100 or so a pop) was simply not in the cards -- like many small businesses we use what are effectively high end desktops as servers.  The durability of some removable magnetic media is attractive, but the accrued cost of sufficient media to do reasonable backups was not encouraging.

The strategy currently in use is a combined approach of of multi-node replication, local caching and conventional backups to inexpensive external disk drives made visible as internal network shares.  All application data, documents and media are distributed internally as network shares through the Microsoft distributed filesystem -- DFS.  These shares are all replicated across two or three different file servers by the OS.  At regularly scheduled times, a conventional backup job runs to copy these shares and other key information to an external USB drive, currently 1TB in size. While inferior to continuous backup schemes used in more expensive storage appliances, the volume shadowing approach implemented by Microsoft to snapshot open files in a consistent state seems to work well -- the only side effects we have observed is that during the snapshot process there is a brief window when the target files may become unavailable to users. It also works better with smaller documents -- DFS replication of very large files (100mb and above) is problematic and probably not appropriate.

Volume shadow copies are used indirectly to effect multi-host file replication through the Microsoft distributed file system. When a network share is added to the dfs name space (available to all network users through the active directory), additional shares on other machines can be added transparently behind the scenes -- and dfs will replicate new and changed files to those shares in accordance with administrator-selected  topologies.  This technology affords two major opportunities for self-destruction -- initial replication mastering and user access choices. When dfs establishes replication for the first time, the administrator has the opportunity to select which of the constituent shares are the replication master -- the others will be adjusted accordingly. So if an existing share 'X' with files is added to a dfs namespace and a new share 'XY' on another machine is added as a replication set member -- the administrator could choose 'XY' as the replication master and transparently delete the existing files on 'X'. Not good. The other opportunity is on the user workstation -- where one can choose which member node in a dfs share to access. Because file propagation can take time and proceeds according to the replication architecture chosen by the administrator, it is possible to update the target rather than the master and break replication.

Microsoft has an additional technology that may work in some environments - Data Protection Manager. The DPM master maintains a backup pool fed through the volume shadow copy snapshot process -- effectively periodic backups of file shares. An add-on client allows end users to select older versions of files for recovery. The 2007 version will also snapshot Exchange files, which makes it much more useful for general purpose backups. We had tested the prior version and are awaiting the 2007 updates.