On Thu, 16 Apr 2003 cgd@broadcom.com wrote:

> At Wed, 16 Apr 2003 23:27:28 +0000 (UTC), "Steve Woodford" wrote:
> > So, what are the main features/benefits of the new pmap?
>
> Have you benchmarked the actual impact on performance?  If so, what
> are the results?
>
> (If not, well, how do you know that things are really faster?  8-)

Heh, as Jason said, the new pmap has been benchmarked on a 600MHz i80321
xscale cpu. Here is a quick summary:

                 L M B E N C H  1 . 9   S U M M A R Y
                 ------------------------------------
                 (Alpha software, do not distribute)

Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
Host                 OS  Mhz null null      open selct sig  sig  fork exec sh
                             call  I/O stat clos       inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ----
evbarm-ne   NetBSD OLD   594  0.9  4.2   37  641 0.06K  2.1    4 4.0K  14K  24K
evbarm-ne   NetBSD NEW   594  0.6  3.6   36  611 0.06K  1.8    3 2.3K  11K  19K

Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host                 OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
                        ctxsw  ctxsw  ctxsw ctxsw  ctxsw   ctxsw   ctxsw
--------- ------------- ----- ------ ------ ------ ------ ------- -------
evbarm-ne   NetBSD OLD   134    237    500   238    499     239     502
evbarm-ne   NetBSD NEW   126    230    493   224    490     224     490

*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host                 OS 2p/0K  Pipe AF     UDP  RPC/   TCP  RPC/ TCP
                        ctxsw       UNIX         UDP         TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
evbarm-ne   NetBSD OLD    134   265  326   600         691       1721
evbarm-ne   NetBSD NEW    126   251  308   567   730   640   895 1585

File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
Host                 OS   0K File      10K File      Mmap    Prot    Page
                        Create Delete Create Delete  Latency Fault   Fault
--------- ------------- ------ ------ ------ ------  ------- -----   -----
evbarm-ne   NetBSD OLD                                  2477          2.1K
evbarm-ne   NetBSD NEW                                  2255          2.0K

*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------
Host                OS  Pipe AF    TCP  File   Mmap  Bcopy  Bcopy  Mem   Mem
                             UNIX      reread reread (libc) (hand) read write
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----
evbarm-ne   NetBSD OLD    36   10   22     26    154     46     47  154    68
evbarm-ne   NetBSD NEW    37   10   23     27    154     48     49  154    68

Memory latencies in nanoseconds - smaller is better
    (WARNING - may not be correct, check graphs)
---------------------------------------------------
Host                 OS   Mhz  L1 $   L2 $    Main mem    Guesses
--------- -------------   ---  ----   ----    --------    -------
evbarm-ne   NetBSD OLD    594     5    181         192    No L2 cache?
evbarm-ne   NetBSD NEW    594     5    181         192    No L2 cache?

There are small variations in the figures between consecutive runs of
lmbench, but the above samples are fairly typical for the development
machine.

Cheers, Steve

-- 

Wasabi Systems Inc. - The NetBSD Company - http://www.wasabisystems.com/