Re: sparc 2 usable interrupt load?

["..I'd rather be coding ASM!" <uridium%deviate.fi@localhost>]

See if you can lay your hands on a Weitek "SPARC Power-UP!" CPU + rom (rom 
is important!) and feed that to it. They're a clock-doubled 80Mhz cpu+fpu 
with onboard cache that specifically on the SS2 and IPX can turn the older 
L2 cache into L3. I've got one and it can so long as it's not swapping and 
not running other apps, idle with around 30% cpu free at wire speed 
thumping qe0 and qe2. The onboard le0 though seems to have much higher int 
service times ..or something causing latency.

Throughput tends to be about 970kb/s in/out of both interfaces.

Regards,
Al.

On Sun, 8 Aug 2010, Aaron J. Grier wrote:

> I've been running 1.6.1 on a trusty sparc2 for years.  it handles
> routing and NAT duties, but after increasing load on my wireless
> segment, it seems to be melting down just passing packets:
>
> le0 at sbus0 slot 0 offset 0xc00000 level 4 (ipl 5): address 08:00:20:11:56:75
> le0: 8 receive buffers, 2 transmit buffers
> le1 at sbus0 slot 1 offset 0xc00000 level 5: address 08:00:20:11:56:75
> le1: 8 receive buffers, 2 transmit buffers
> le2 at sbus0 slot 2 offset 0xc00000 level 5: address 08:00:20:11:56:75
> le2: 8 receive buffers, 2 transmit buffers
>
> vmstat -i:
>
> interrupt                                     total     rate
> lev1                                      404051709       36
> lev3                                       63835545        5
> lev5                                      942141929       85
> lev6                                           1507        0
> clock                                    1097372546       99
> lev12                                         34030        0
> prof                                     1097186333       99
> zs0 intr                                      34030        0
> esp0 intr                                 280432982       25
> Total                                    3885090611      354
>
> this with 98+% interrupt time according to top.  interactive shell
> through an ssh session is extremely laggy.  the bulk of traffic flows
> are between le2 and le0.
>
> I realize this is only a 40MHz machine, but I would have thought it able
> to pass 10mbit traffic at close to wire speed.  am I mistaken?  have I
> really hit the performance limit?  (could an ISA 486SLC do better?)

--
 --
 Al Boyanich
 adb -w -P "world> " -k /dev/meta/galaxy/ksyms /dev/god/brain