Port-sparc64 archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

wm(4) on sparc64 (was: sk(4) on sparc64)



On Wed, Oct 10, 2007 at 05:06:07AM +0200, Holger Weiss wrote:
> * Chris Ross <cross+netbsd%distal.com@localhost> [2007-09-23 22:44]:
> >   I wanted to report that this interface works just fine on my sparc64.  At
> > least at 100Mbit, full-duplex, which is the switch I have it hooked to.
> 
> FTR, sk(4) negotiates 1000Base-T (full-duplex) just fine on my Ultra 80,
> though I'm not able to actually pipe more than ~90 Mb/s through the NIC
> on that system.

Now that PCI cards finally work in my ULTRA60 because Julian fixed the
interrupt handling I could benchmark an Intel PRO1000MT card:

wm0 at pci1 dev 1 function 0: Intel i82540EM 1000BASE-T Ethernet, rev. 2
wm0: interrupting at ivec 0
wm0: 32-bit 66MHz PCI bus
wm0: 64 word (6 address bits) MicroWire EEPROM
wm0: Ethernet address 00:07:e9:XX:XX:XX
makphy0 at wm0 phy 1: Marvell 88E1011 Gigabit PHY, rev. 3
makphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT, 
1000baseT-FDX, auto

The receive performance is very bad:

ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001, sockbufsize=131072  
tcp
ttcp-r: socket
ttcp-r: rcvbuf
ttcp-r: accept from 81.187.181.119
ttcp-r: 268435456 bytes in 25.85 real seconds = 10140.71 KB/sec +++
ttcp-r: 32772 I/O calls, msec/call = 0.81, calls/sec = 1267.74
ttcp-r: 0.0user 3.2sys 0:25real 12% 0i+0d 0maxrss 0+1pf 0+0csw

TSO seems to help with the send performance:

ttcp-t: buflen=32768, nbuf=8192, align=16384/0, port=5001, sockbufsize=131072  
tcp  -> colwyn
ttcp-t: socket
ttcp-t: sndbuf
ttcp-t: connect
ttcp-t: 268435456 bytes in 12.12 real seconds = 21624.37 KB/sec +++
ttcp-t: 8192 I/O calls, msec/call = 1.52, calls/sec = 675.76
ttcp-t: 0.0user 1.1sys 0:12real 9% 0i+0d 0maxrss 0+32809pf 0+0csw

This is definitely disappointing, Solaris 10 achieves at least four
times the receive performance and twice the send performance on
similar hardware. It does however utilize SMP to do that. So there
is hope for improvement.

        Kind regards

-- 
Matthias Scheler                                  http://zhadum.org.uk/



Home | Main Index | Thread Index | Old Index