Check out the new USENIX Web site. next up previous
Next: 3 Security Up: 2 The simplified scheme Previous: 2.5 Handling downtime


2.6 Resizing the amount of backup space

Consider a computer currently swapping $s$ blocks with each of $n=k{+}m$ partners. In return for $n{\times}s$ blocks of local disk, it has access to a logical disk of size $k{\times}s$ blocks. If it needs additional logical-disk space, it can either add more partners swapping $s$ blocks each (presumedly maintaining a similar ratio of $k$ and $m$) or switch to $n$ new partners willing to swap $s'>s$ blocks each. Adding partners increases the amount of overhead due to per-partner costs (especially under the full scheme where we must periodically check on each partner), but requires issuing a new current partner list less often.

The same methods run in reverse can be used to shrink the logical disk. Partners holding redundancy blocks can also be added or removed to adjust the reliability level. Most of these changes require moving data around to maintain a sequential image (i.e., adding partners adds blocks to every stripe row, rather than just adding a bunch of blocks at the end of the disk). By using the master block and version numbers, this can be done using no extra space in a restartable way with restoration always possible.


next up previous
Next: 3 Security Up: 2 The simplified scheme Previous: 2.5 Handling downtime
Mark Lillibridge 2003-04-07