| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 60.1 | L3 cache is like all other external caches | UNIFIX::HARRIS | Juggling has its ups and downs | Tue Mar 11 1997 12:45 | 16 |
| Are you talking about some Alpha Chip implementations which have an L1
(level 1) and L2 cache built into the chip and therefore the external
cache is called L3?
If so, then an L3 cache works just like every other external cache. It
is built using memory chips that have a much faster access time then
main memory, it is attached to the CPU via a fast interface, and it is
much larger than the L1 & L2 caches inside the CPU chip. I think the
L1 cache is 8K instruction, 8K data, the L2 cache is 96K and handles
both instruction and data, and the L3 caches tend to be a megabyte or
more and vary in size depending on the system (servers get more,
workstations get less). While I'm not 100% sure of the sizes, I'm
fairly sure I'm in the ball park. Someone will correct me if I'm wrong
(so I expect several corrections :-)
Bob Harris
|
| 60.2 | See Rawhide 79.* | XDELTA::HOFFMAN | Steve, OpenVMS Engineering | Tue Mar 11 1997 15:28 | 16 |
|
A cache brings more commonly accessed data or code from memory that
is (relatively) slow into memory that is (relatively) fast, and to
prevent the processor from being delayed by writing back data into
(relatively) slow memory.
There are various caching schemes, and various caching implementations
on various Alpha microprocessors and on VAX processors and microprocessors.
As for specifics, I suspect this is a Rawhide-series system involved
(an AlphaServer 4000 or 4100 series system), and the customer will
likely find the the board-level (B-cache) cache advantageous under
various common operations. See the Rawhide (MVBLAB::ALPHASERVER_4100)
notes conference notes 79.*, among others, for some discussion of the
advantages of the B-cache.
|
| 60.3 | a rare case | STAR::jacobi.zko.dec.com::jacobi | Paul A. Jacobi - OpenVMS Systems Group | Tue Mar 11 1997 16:55 | 6 |
|
There's also an interesting discussion in MVBLAB::ALPHASERVER_4100 about a
rare case where the 3rd-level cache actually degrades the performance.
-Paul
|
| 60.4 | Cache or cash? | BBPBV1::WALLACE | john wallace @ bbp. +44 860 675093 | Wed Mar 12 1997 03:45 | 23 |
| In industry-standard salesman's terms :-)
Your level 1 cache is like your wallet. That's where important
frequently used things go to and from, and it's designed for very quick
access. But it's not very big and although there's usually enough in
there, it helps to have other places to go if you happen to need
something that isn't in L1.
The L2 cache is like your day to day checking account. It's bigger than
the L1 cache (wallet) but takes a little while to get things out of L2
to where you can work with them. Not too long, but a little while.
Main memory is like a fixed-notice savings account. You can request
something out of it but it will be a good while before you can do
anything useful with it. It helps to plan ahead. In the interim period,
if you've nothing else to work with, you just have to sit back and
relax while the new input arrives.
What's L3 cache in this model? Don't know, except I've not got one...
just fit it in between L2 and main memory.
regards
john
|
| 60.5 | Thanks | VAXRIO::CUPOLILLO | | Wed Mar 12 1997 14:38 | 6 |
|
Thanks, for the replies, I will check the notes on RAWHIDE. My main
concern is about how the performance can be improved adding one more
level of cache.
Cupol
|