I was happy /w the re driver too until 4.4 (I think my previous
firewall/samba share server was 4.2)
I did mention in my original post I was watching systat vmstat, during
the post I mentioned I was watching hard drive kbyte writes. When I
decided to run out and buy the em I noticed my re was doing 4 to 6k+
interrupt requests a second. and I was "tweaked" samba it rose even
higher via systat vmstat. I didn't mention the interrupt storm because
I no longer want to fuss the system or argue /w tico or mess with
duplex settings as it was all pointless -  as I saw it already had the
solution to my problem in hand for 36+cdn $: the new em card.

Am I happy about the change? hell ya, my doorstop compaq pentium 866
writes to the 1GB WD hard drive between 18000+k every systat vmstat
blip. I ain't complaining now.

I only mention it here because the original poster of the thread
didn't mention how his network was configured. If his nvidia+8169s phy
is his samba interface I'm left to wonder if he is seeing something
similar to what I was experiencing. And like a typical end user I was
after results, thus instead of further testing & further listerv
followup i threw money at it to make it work to a point where I was
satisfied..

> rl is pretty different to re. I'm fairly happy with re(4) considering
> how cheap they are.
>
> re -> bge

Is that loopback or between two boxes with a switch in the middle?

> re-box$ tcpbench bge-box
>     pid   elapsed_ms          bytes         Mbps
>   15569         1030       69931144      543.155
>   15569         2024       67539536      544.125
>   15569         3027       68251072      544.375
>   15569         4020       67548992      544.201
>   15569         5024       68086608      543.064
>   15569         6027       67858352      541.243
>   15569         7018       67436728      544.943
>   15569         8022       68277960      544.590
>   15569         9025       68331152      545.014
>
> of course the more expensive NICs are better. bge -> re
>
> bge-box$ tcpbench re-box
>     pid   elapsed_ms          bytes         Mbps
>   31564         1007      110421616      877.232
>   31564         2009      104687432      836.663
>   31564         3007      104840696      840.406
>   31564         4007      109819488      879.435
>   31564         5007      116304656      931.369
>   31564         6007      115594504      925.682
>   31564         7007      116234288      930.805
>   31564         8007      116101120      929.739
>   31564         9006      116120392      929.893
>   31564        10006      116213104      930.635
>   31564        11006      115465648      924.650
>   31564        12006      116225744      929.806
>   31564        13006      116123152      929.915
>
> (single tcp stream, opteron 146 on a supermicro aplus board
> running i386 - for anyone interested, CWM on the bge was peaking
> at 46).
>
> re0 at pci1 dev 0 function 0 "Realtek 8168" rev 0x01: RTL8168 2 (0x3800),
apic 2 int 16 (irq 11), address 00:30:18:a0:6a:f6
> rgephy0 at re0 phy 7: RTL8169S/8110S PHY, rev. 2
>
> bge0 at pci2 dev 3 function 0 "Broadcom BCM5704C" rev 0x10, BCM5704 B0
(0x2100): apic 2 int 8 (irq 5), address 00:30:48:58:86:40
> brgphy0 at bge0 phy 1: BCM5704 10/100/1000baseT PHY, rev. 0
>
>> My ports/misc thread:
>> http://marc.info/?l=openbsd-ports&m=122703016321404&w=2
>> http://marc.info/?l=openbsd-misc&m=122719611210846&w=2

<!-snip-!>

Cheers & good day.

Reply via email to