HP Vectra XU 6/XXX HP Vectra XU 6/XXX - Guide to Optimization Performance - Page 38

Dedicated Level-Two Cache Bus, Cache Coherency, The MESI Protocol

Get HP Vectra XU 6/XXX PDF manuals and user guides View all HP Vectra XU 6/XXX manuals
Add to My Manuals
Save this manual to your list of manuals

Page 38 highlights

Dedicated Level-Two Cache Bus The Pentium Pro increases the performance of its level-two cache by accessing it over a dedicated, 64-bit internal bus. This means that accesses to the cache will cause no contention with any operations on the local bus. It also ensures that the level-two cache is able to operate at the full internal clock speed of the processor. Cache Coherency One important implication of using cache memories is that duplicate copies of data exist in different parts of your PC. With duplicate copies of the same data, a mechanism must be used to ensure coherency between the copies. This is particularly true for dual processor configurations, where duplications can exist not only between a processor cache and memory, but also between the caches in the two processors. The Pentium Pro's level-two data cache is a copy-back cache, meaning that if the processor writes data that produces a cache hit, then that data will be written directly in the level-two cache. The cache will then copy the data back to memory at a later time. Until the cache copies the data back to memory, the copy of the data held in memory is no longer valid. If another device in your PC requests this data from memory, the Pentium Pro processor must be able to intercept the request so that it can supply the correct version of the data. To meet this requirement, the Pentium Pro processor uses two cache coherency mechanisms: the MESI (Modified, Exclusive, Shared, Invalid) protocol; and bus snooping. The MESI Protocol The MESI protocol defines the status of data in the cache using four states: • Modified - the data is valid, but different to that in memory (the data in memory is invalid) • Exclusive - the data is valid and contained in one processor cache only

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
Dedicated Level-Two Cache Bus
The Pentium Pro increases the performance of its level-two cache by accessing it over a
dedicated, 64-bit internal bus. This means that accesses to the cache will cause no contention
with any operations on the local bus. It also ensures that the level-two cache is able to operate
at the full internal clock speed of the processor.
Cache Coherency
One important implication of using cache memories is that duplicate copies of data exist in
different parts of your PC. With duplicate copies of the same data, a mechanism must be used
to ensure coherency between the copies. This is particularly true for dual processor
configurations, where duplications can exist not only between a processor cache and memory,
but also between the caches in the two processors.
The Pentium Pro’s level-two data cache is a copy-back cache, meaning that if the processor
writes data that produces a cache hit, then that data will be written directly in the level-two
cache. The cache will then copy the data back to memory at a later time. Until the cache
copies the data back to memory, the copy of the data held in memory is no longer valid.
If another device in your PC requests this data from memory, the Pentium Pro processor must
be able to intercept the request so that it can supply the correct version of the data. To meet
this requirement, the Pentium Pro processor uses two cache coherency mechanisms: the MESI
(Modified, Exclusive, Shared, Invalid) protocol; and bus snooping.
The MESI Protocol
The MESI protocol defines the status of data in the cache using four states:
•
Modified — the data is valid, but different to that in memory (the data in memory is
invalid)
•
Exclusive — the data is valid and contained in one processor cache only