Andrea Arcangeli (andrea@e-mind.com)
Wed, 16 Dec 1998 13:06:46 +0100 (CET)
On Wed, 16 Dec 1998, Philip Blundell wrote:
>No; if your printer is very fast then the right thing to do is just to reduce
>the udelay() inside the polling loop.
OK.
>Imagine your printer takes 5us to process a byte for example. If you never
If my printer is busy for more than *xusec it means that it will be busy
for some seconds then and this is mainly always true.
The point is that if your machine need to print fast as first thing you
_must_ use polling and not irq driven printing, and this is true now and
will be true forever! Everything out there would be faster using polling.
My old IDE disk was faster without dma following this discussion. The
point of irq driven printing (or dma) is to decrease the CPU load (for DMA
also improve performances btw) and to give it a sense we __must__ sleep
waiting for the irq. I can agree that some printers need a bit more of
very very fast polling because their *xusec could be longer then mine...
Probably we should account something like the RTO (mean variation) of the
TCP/IP in the pardevice structure to know the mean time after which we
must consider the device extra slow and so sleep on the irq safely to have
not broken performance in the fast case...
>For most people, printing at a reasonable speed is more important than
>reducing CPU loading while the port is in use.
Ok, but remeber that the printing is per default a background job so I
don' t want to excess to the other point.
I sure agree with fast poll for a bit even if we are irq driven but I
would like to still see many of mine "Hoho the irq make sense" messagess
as now... also because the printing here is equally fast than the polled
printing and the CPU is very ligher used. I reach ~100000 irq per sec as
peak in postscript printing.
Andrea Arcangeli
-- To unsubscribe, send mail to: linux-parport-request@torque.net --
-- with the single word "unsubscribe" in the body of the message. --
This archive was generated by hypermail 2.0b3 on Wed 30 Dec 1998 - 10:18:55 EST