Re: Re: Re: Re: Question: Rx Ring Buffers in the rtl8139 module
Donald Becker <[email protected]>
| Newsgroups | gmane.linux.drivers.realtek.devel |
|---|---|
| Message-ID | <[email protected]> |
On Fri, 30 Aug 2002 [email protected] wrote: > Donald Becker <[email protected]> schrieb am 30.08.02 20:13:19: > > On Fri, 30 Aug 2002 [email protected] wrote: > > > > > 8139too gives in the order of 10.000 errors per sec under load, while > > > rtl8139 is better and has (only) a few 100 errors per sec. > > > > > > Is there anything I can do about that? > > > > Turn on flow control, if you have a flow control switch. > > Tune the PCI parameters. > > > > What are the errors? Dropped packets? Or CRC errors? > > dropped packet (ifconfig states the same sumber for dropped as > for overruns). There are onyl Rx no Tx errors. What are the errors reported in /proc/net/dev? > Also nfs hangs after very few accesses: mounting a remote fs > is ok, the first ls /mnt is often ok (but not always) and one of > the next ls /mnt/somedir will hang. > > > > It's an onboard NIC on a laptop. > > > > Which might have a very slow PCI system or processor.. > > not really: it's a athlon 1200 mhz mobile. should be enough for a single > TCP connection (my testcase is scp to the laptop). It could still have a very slow PCI system. The chip requires at least a 25Mhz PCI clock, and reasonable bandwidth to main memory. -- Donald Becker [email protected] Scyld Computing Corporation http://www.scyld.com 410 Severn Ave. Suite 210 Second Generation Beowulf Clusters Annapolis MD 21403 410-990-9993