Re: CFT: re(4)

Pyun YongHyeon <pyunyh_at_gmail.com> wrote:

> I've committed a fix for bus_dma(9) bug which resulted in poor Tx
> performance on TSO enabled re(4) driver. With the fix and revised
> re(4) I got more sane performance on re(4). Because there are too many
> hardwares that rely on re(4) I'd like to hear any success or failure
> reports before revised re(4) hits the tree.

With:

fk_at_africanqueen ~ $pciconf -lv | grep -A 4 ^re 
re0_at_pci0:9:0:   class=0x020000 card=0x816910ec chip=0x816910ec rev=0x10 hdr=0x00
    vendor     = 'Realtek Semiconductor'
    device     = 'RTL8110SB Single-Chip Gigabit LOM Ethernet Controller'
    class      = network
    subclass   = ethernet

and netserver running on the host with re(4),
and netperf running on a system with em(4)
and FreeBSD 6.2-STABLE, I got:

With TSO4 disabled:

Stock re(4):

 65536  32768  32768    10.00     461.63   
 65536  32768  32768    10.01     459.43   
 65536  32768  32768    10.01     460.71 

Patched re(4):

 65536  32768  32768    10.00     459.92   
 65536  32768  32768    10.01     461.14   
 65536  32768  32768    10.01     460.54

With TSO4 enabled:

Stock re(4):

 65536  32768  32768    10.00     460.12   
 65536  32768  32768    10.00     460.32   
 65536  32768  32768    10.01     461.89

Patched re(4):

 65536  32768  32768    10.01     459.97
 65536  32768  32768    10.01     459.27   
 65536  32768  32768    10.01     459.60

Changing the position of netperf and netserver
(netperf now running on the host with re(4)):

With TSO4 disabled:

Stock re(4):

 65536  32768  32768    10.00     326.10   
 65536  32768  32768    10.00     325.81   
 65536  32768  32768    10.00     325.75

Patched re(4):

 65536  32768  32768    10.00     325.89   
 65536  32768  32768    10.00     327.28   
 65536  32768  32768    10.00     328.06

With TSO4 enabled:

Stock re(4):

 65536  32768  32768    10.00     326.64   
 65536  32768  32768    10.00     326.46   
 65536  32768  32768    10.00     326.41

Patched re(4):

 65536  32768  32768    10.00     328.22   
 65536  32768  32768    10.00     328.03   
 65536  32768  32768    10.00     328.41

During the tests with the stock re(4) I got
"re0: can't map defragmented mbuf (error 0)"
four times. I didn't get any warnings with
the patched one.

em(4) was running with standard options all the time.

re0 is compiled into the kernel and I had
to reboot between the test triples.

Fabian