[cobalt-users] kernel update and bus speed explained

[Jeff Lovell <jlovell@xxxxxxx>]

The following is details from one of our kernel developers
as to the recent discussion of bus speeds after the latest
kernel update.

Jeff

----- Forwarded message from Tim Hockin -----

Hey all, I just want to respond and clarify, so no one gets worried.


> Does anyone know what bus speed the RaQ3's are *suppose* to be running? I

This is a bit of a trick question - there are multiple busses in the
system.  The PCI bus runs at 33 MHz, 32 bits (PCI standard).  The
Front-side bus (CPU-memory) runs at 100 MHZ (PC100 DIMMs).

> looking around, and couldn't find anything out of the ordinary, but it just
> seems like it?s a bit slower after installing this kernel update. So I went

I sincerely doubt it is slower.  Our internal numbers show it marginally
faster or the same for every benchmark.  I encourage you to run something
like lmbench if you think there is a performance issue.  If there is an
anomaly, please let us know!

> 33MHz... Now I realize little blue is terribly underpowered, but I was certain
> they were running at 66MHz and NOT 33MHz prior to this update. That would

This is a red herring. see below :)

> kernel: Uniform Multi-Platform E-IDE driver Revision: 6.30
> kernel: ide: Assuming 33MHz system bus speed for PIO modes; override with
> idebus=xx
> kernel: ALI15X3: IDE controller on PCI bus 00 dev 78
> kernel: ALI15X3: chipset revision 193
> kernel: ALI15X3: 100%% native mode on irq 14
> kernel:     ide0: BM-DMA at 0xf000-0xf007, BIOS settings: had:DMA, hdb:DMA
> kernel:     ide1: BM-DMA at 0xf008-0xf00f, BIOS settings: hdc:DMA, hdd:DMA
> kernel: had: ST310212A, ATA DISK drive
> kernel: ide: Assuming 33MHz system bus speed for PIO modes; override with
> idebus=xx
> kernel: ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
> kernel: had: ST310212A, 9768MB w/512kB Cache, CHS=19846/16/63, UDMA(33)

Notice above that both IDE0 and IDE1 are BM-DMA (that is BusMastering
Direct Memory Access).  That is good :).  Notice also that the drives are
set to DMA.  That is also good.

The message about 33MHz means deserves a bit of background.  In the olden
days, IDE drives used Programmed IO (PIO).  This meant the CPU had to
request blocks, then copy them to memory, then request more blocks, then
copy them to memory.  You can see how this abuses the CPU while you are
doing disk IO.  Modern drives use DMA.  The CPU tells the drive what blocks
to copy to memory, and where to put them.  Then it does something else
while the drive does it's work.  When the drive is done, an IRQ is
delivered.  This yields much better throughput and CPU usage.

The message above is merely indicating that the system bus speed (the bus
on which the IDE controller lives: PCI) is assumed to be 33MHz.  This is
absolutely a correct assumption.  However, the message also shows that the
need for that info is for PIO modes, which we aren't using - we're doing
DMA.

I hope that clarifies some.  If there are further questions,I'd be happy to
clarify.

Tim
-- 
Tim Hockin
Systems Software Engineer
Sun Microsystems, Cobalt Server Appliances
thockin@xxxxxxx

----- End forwarded message -----