IBM 610 workstation computer 3402

KR Williams

snip

Now we can have a sensible debate.

The lack of fast memory was the reason virtual storage systems were invented. Each virtual storage block has a pointer is a table somewhere. In some machines the hardware controls this table in other architectures only the software know that it exists. The entries in the table tend to include the area's virtual address, actual address i.e. start location, size and access rights. There is frequently a block "dirty" bit to show that the block needs writing to disk.

Changing a single bit normally results in the entire say 512 byte block being written to the hard disk (4k byte pages on Windows XT). Capability machines like the Plessey S250 can support variable sized blocks. Adding an extra column to this table to tell the the CPU to always write off chip is not difficult.

When page faults occur it is quite normal for ram (and rom) areas to be swapped out onto disk. The virtual memory part of the operating system could be extended to handle the on chip memory.

Data transfers between ran and disk controller are sometimes handled by DMA across the bus and sometimes by copying the words through the CPU's registers. Either way there is a lot of traffic on the bottleneck known as the bus. To permit this to work current cache memories have to write updates into main memory several times a second since DMA chips have difficulties talking to L1 cache on microprocessor chips. Frequently updated locations are written lots of times where as only the last write is needed. Replace the cache with on chip ram and the data can stay on chip until the area is reclaimed, possibly minutes later.

Andrew Swallow