| It is difficult to characterise the Oracle7 on ncube2 as it mixed both shared-disk and shared-nothing architectures features by using an additonal subcube as a giga-cache which lies between ncube I/O system (including I/O nodes, multiple I/O channels and disks drives) and the compute-node subcube(s). |
| Before data is read into the buffer in compute-processor, they first are cached in a Giga-cache node. |
| So if you look just at the compute-subcube <--> Giga-cache subcube, it looks likea shared-nothing system, |
| But if you look at the whole compute-subcube <--> Giga-cache subcube <--> disk-arrays, it looks like a shared-disk system. |
| Strictly speaking and compared to DB2 on SP2, it is not a shared-nothing system as the data-placement on disk-arrays has little to do with how query processing is paralleliazed or decomposed. |
| This is also the reason why we found data partition schemes have less I/O performance impact for Oracle7 on nucbe than that on SP2 (the latter is a shared-disk system). |