[H-GEN] Time loss on copy

Martin Pool mbp at linuxcare.com.au
Tue Jun 13 00:43:11 EDT 2000

[ Humbug *General* list - semi-serious discussions about Humbug and ]
[ Unix-related topics.  Please observe the list's charter.          ]

On Tue, Jun 13, 2000 at 01:48:39PM +1000, Frank Brand wrote:

> Has to do with the timing signals that the RealTek networking card are
> sending to the PC.

What timing signals are they?  Does he mean interrupts, or is he
implying that the NIC hooks the clock interrupt.

Anyhow, is it RealTek or NE2000?  I think they are different cards.

> The PC attempts to give the Network card all available
> resources during a large intensive copy and it 'assumes' that the timing
> signals will be in sync with what the BIOS reports back as the
> correct time.

Where does the BIOS come into it?

> There is an option for a NE2000 compatible card that keeps it in sync, but
> the downside is that the copy will take longer.
> In a sentence - The network cards clock is faster than the PC's.

If it is an RT card, then I can imagine that this comment from the
source would indicate why the kernel is slow to process interrupts:

 * III. Driver operation
 * IIIa. Rx Ring buffers
 * The receive unit uses a single linear ring buffer rather than the more
 * common (and more efficient) descriptor-based architecture.  Incoming frames
 * are sequentially stored into the Rx region, and the host copies them into
 * skbuffs.
 * Comment: While it is theoretically possible to process many frames in place,
 * any delay in Rx processing would cause us to drop frames.  More importantly,
 * the Linux protocol stack is not designed to operate in this manner.
 * IIIb. Tx operation
 * The RTL8129 uses a fixed set of four Tx descriptors in register space.
 * In a stunningly bad design choice, Tx frames must be 32 bit aligned.  Linux
 * aligns the IP header on word boundaries, and 14 byte ethernet header means
 * that almost all frames will need to be copied to an alignment buffer.
Martin Pool, Linuxcare, Inc.
+61 2 6262 8990
mbp at linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

* This is list (humbug) general handled by majordomo at lists.humbug.org.au .
* Postings to this list are only accepted from subscribed addresses of
* lists 'general' or 'general-post'.

More information about the General mailing list