[RC5] AMD K6-233, Win32 & DOS client strangeness...

Art Coates Jr. axis at aye.net
Wed Aug 26 03:03:52 EDT 1998


    I have a strange problem with the current DOS (v2.7100.4160) and
Win95 (v2.7100.4160) rc5des clients.  I've been using them and the
current Linux (S.u.S.E. 5.2, 2.0.35 kernel compiled by myself) clients
for about 11 days now, and of course playing with all the settings to
find the appropriate settings for my computer and my own personal
tastes.  I built this computer myself and it consists of a standard AMD
K6-233 (66mhz bus setting, 3.5X multiplier), PcChips M571 motherboard
with TXProII chip-set. Both Win95 and Linux are installed dual-boot via
LILO.  I have no unusual problems or anomalies in Linux. However: in DOS
or Win95 using the default settings I have noticed that I do not achieve
a key rate as high as I do in Linux.  While this may be dismissed as the
difference of the operating systems involved, I decided to investigate
further.  I tried different settings in both the Win95 and DOS clients
and found out a setting that gave me an equal (within 200 keys/sec) rate
as I would achieve running Linux on the same system.  Rather than using
the default auto detect or K6 processor settings, I use the "Pentium,
Pentium MMX, Cyrix486/5x86/MediaGX, AMD 486" -settings in DOS/Win95.
This improves my key rate.  Here are copies of my log, from when the
system was completely idle in (pressing F8 on boot, then selecting:
Command Prompt Only) DOS:


DOS client, default (-1) processor setting (correctly detects AMDK6):
---------------------------------------------------------------------
RC5DES Client v2.7100.416 started.
Using axis at aye.net as email address.

DES blocks are currently unavailable; Falling back to RC5
[Aug 21 12:10:42 GMT] Loaded RC5 1*2^30 block 70CF02ED:40000000
[Aug 21 12:10:43 GMT] 49 RC5 blocks remain in file buff-in.rc5
[Aug 21 12:10:43 GMT] 0 RC5 blocks are in file buff-out.rc5

[Aug 21 12:57:16 GMT] Completed RC5 block 70CF02ED:40000000 (1073741824
keys)
[Aug 21 12:57:16 GMT] 0.00:46:33.84 - [384,324.73 keys/sec]
[Aug 21 12:57:16 GMT] Summary: 1 RC5 block 0.00:46:33.84 - [384.32
kkeys/s]
DES blocks are currently unavailable; Falling back to RC5
[Aug 21 12:57:17 GMT] Loaded RC5 1*2^30 block 70CF02EC:C0000000
[Aug 21 12:57:17 GMT] 48 RC5 blocks remain in file buff-in.rc5
[Aug 21 12:57:17 GMT] 1 RC5 block is in file buff-out.rc5


DOS client using (2) Pentium, Pentium MMX setting:
-----------------------------------------------------
RC5DES Client v2.7100.416 started.
Using axis at aye.net as email address.

DES blocks are currently unavailable; Falling back to RC5
[Aug 25 13:24:23 GMT] Loaded RC5 1*2^30 block 70CF01A8:C0000000 (72.48%
done)
[Aug 25 13:24:23 GMT] 8 RC5 blocks remain in file buff-in.rc5
[Aug 25 13:24:23 GMT] 0 RC5 blocks are in file buff-out.rc5

[Aug 25 13:37:05 GMT] Completed RC5 block 70CF01A8:C0000000 (1073741824
keys)
[Aug 25 13:37:05 GMT] 0.00:12:41.71 - [387,807.84 keys/sec]
[Aug 25 13:37:05 GMT] Summary: 1 RC5 block 0.00:12:41.71 - [387.80
kkeys/s]
DES blocks are currently unavailable; Falling back to RC5
[Aug 25 13:37:05 GMT] Loaded RC5 1*2^30 block 70CF01A8:80000000
[Aug 25 13:37:05 GMT] 7 RC5 blocks remain in file buff-in.rc5
[Aug 25 13:37:05 GMT] 1 RC5 block is in file buff-out.rc5


-These results are repeatable, day after day and boot after boot.  Now,
when the system is idle in Win95, DOS or Linux, the key rates are almost
key-for-key identical to these results.  Note that in Linux, the default
(-1) setting is used to achieve this key rate.  No other setting in
Linux is as fast as the correct K6 or auto-detect setting, this is
unlike the Pentium, Pentium MMX setting required for both the DOS and
Win95 clients on my system..  I am not familiar with processor design or
rc5des client code, but my guess would be that the processor cores for
K6 are not yet fully optimized in the DOS and Win95 clients.  Again, I
am not an expert, and for lack of a comparable system to mine I could
accept the fact that it is my specific hardware/software that causes
this.  Maybe a 3483 key/sec improvement isn't enough to worry about.
For AMDK6 users that may want to try this for a faster key rate or
rc5des client coders that would want to take a look and see if there is
a small improvement that may be made in the Win95/DOS K6 code -I have
submitted this.  Either way, I hope this helps someone, maybe even
myself (maybe I should be getting even faster than 387,807 keys/sec?).
;)



-Art Coates Jr.



--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list