Re: zero-copy TCP fileserving

Richard Gooch (rgooch@atnf.csiro.au)
Fri, 4 Jun 1999 22:57:17 +1000

Messages sorted by: [ date ][ thread ][ subject ][ author ]
Next message: Dave Jones.: ".ident's taking up kernel space"
Previous message: Alan Cox: "Re: zero-copy TCP fileserving"
In reply to: Alan Cox: "Re: zero-copy TCP fileserving"

Alan Cox writes:
> > > As most (if not all) network drivers do load the entire frame onto the
> > > card before beginning transmission, this is a possibility.
> >
> > Linus has pointed out that good cards fetch the data from memory as it
> > is transmitted (is this true?) so I'll eat my words before everyone else
> > does. :-)
>
> Often it isnt true. They send one frame while fetching the next from
> main memory. This avoids tons of crap packets caused by PCI bus
> latency and the like but has almost identical performance under real
> I/O conditions

So then they can usefully do the checksum in hardware, right? That
will help keep the CPU from polluting it's caches (provided we're
using sendfile() or some such).

Regards,

Richard....

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

Next message: Dave Jones.: ".ident's taking up kernel space"
Previous message: Alan Cox: "Re: zero-copy TCP fileserving"
In reply to: Alan Cox: "Re: zero-copy TCP fileserving"