second block of the burst begins transferring before the first block is completed,
and so on for the third and fourth blocks. In terms of speed, pipelined burst
(PLB) cache is slower on its first block than standard synchronous cache because
of the time required to set up the “pipe,” averaging bursts of 3-1-1-1 on systems
with bus speeds of up to 100MHz. This is the caching technology used on most
Pentium-class motherboards.Caching Write Policies
ThedataincacheispassedtoandreceivedbackfromtheCPU.Itissafetoassumethatthe
data the CPU passes back has been updated or changed in some form. If the data in cache
has been changed, it is also a safe assumption that the user wants to save the data back to
the hard disk. There is no direct logical connection between cache memory and the hard
disk. Therefore, some policy must be in effect on how data gets updated in RAM, so it can
be eventually written back to the hard disk. There is also a need to kee pthe data in RAM
anditsmirrorincachesynchronizedtoavoidpassingabadversionofthedatatotheCPU
or hard disk.Cachingwritepoliciesgovern these actions to ensure that the data mirrored in
cache and RAM stays in sync.
There are two basic cache write policies used to control when data in cache is written
back to main memory: Write-back cache If any of the data mirrored in cache is updated in RAM,
only the line affected is updated in cache. When data that has been updated in
cache by the CPU is cleared, the changed portion of the data is then written
back to RAM. This policy saves write cycles to memory, which are time and
cycle consuming. Write-back is better than write-through, in most cases, which
is why it is the most common.
Write-through cache Anytime data held in cache is modified, it is immediately
written to both cache and main memory. This caching policy is simpler to
implement and ensures that the cache is never out of sync with main memory.
However, because it competes for clock cycles, it can contribute to slower
system performance on a very active PC.Nonblocking Cache
Another characteristic of caching systems is that they can be blocking or nonblocking. A
blocking cachesystem handles only one request at a time. This can create performance
problems, especially in the event of a cache miss. While the requested data is trans-
ferred from main memory, the cache is blocked and must wait for the transfer from
RAM to finish. Anonblockingcache, also calledtransactionalcache, sets aside requests for
data not in cache and works on other transactions while the uncached data is trans-
ferred from main memory.(^164) PC Hardware: A Beginner’s Guide