X-Surf II, what a scam...

[patrik]

@Ilwrath:

What results do you get if you benchmark it with ttcp and what cpucard is your amiga equipped with?.. and just for the sake of asking, have you tried it with AmiTCP/Genesis?

Amiga version of ttcp
Windows version of ttcp
Sourcecode for ttcp


/Patrik

[patrik]

Hi everyone!

In the light of this thread I took the opportunity and benchmarked a friends X-Surf 1, mounted in his A3000 with ttcp and here are the results:

Amiga = Amiga A3000 030@25MHz/X-Surf 1/AmigaOS3.1, AmiTCP4.3
PC = Dell Latitude CPi PII@300MHz/Netgear FA411/Linux 2.4.25
Mac = IBook G3-750FX@800MHz/Builtin NIC/MacOS 10.3.3

Switch = HP Procurve 408

TCP: Amiga->PC
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> 10.95.1.73
ttcp-t: 16777216 bytes in 44.04 real seconds = 372.03 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 22.02, calls/sec = 46.50
ttcp-t: 0:44real
TCP: PC->Amiga
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: 16777216 bytes in 32.56 real seconds = 503.19 KB/sec +++
ttcp-r: 2302 I/O calls, msec/call = 14.48, calls/sec = 70.70
ttcp-r: 0:32real
UDP: Amiga->PC
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  udp  -> 10.95.1.73
ttcp-t: 16777216 bytes in 28.40 real seconds = 576.90 KB/sec +++
ttcp-t: 2054 I/O calls, msec/call = 14.16, calls/sec = 72.32
ttcp-t: 0:28real
TCP: Amiga->Mac
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> 10.95.1.10
ttcp-t: 16777216 bytes in 43.18 real seconds = 379.43 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 21.59, calls/sec = 47.43
ttcp-t: 0:43real
TCP: Mac->Amiga
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: 16777216 bytes in 30.36 real seconds = 539.66 KB/sec +++
ttcp-r: 2308 I/O calls, msec/call = 13.47, calls/sec = 76.02
ttcp-r: 0:30real
UDP: Amiga->Mac
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  udp  -> 10.95.1.10
ttcp-t: 16777216 bytes in 30.06 real seconds = 545.04 KB/sec +++
ttcp-t: 2054 I/O calls, msec/call = 14.99, calls/sec = 68.33
ttcp-t: 0:30real

Switch = DLink DES-1008D

TCP: Amiga->PC
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> 10.95.1.73
ttcp-t: 16777216 bytes in 44.10 real seconds = 371.52 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 22.05, calls/sec = 46.44
ttcp-t: 0:44real
TCP: PC->Amiga
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: 16777216 bytes in 32.54 real seconds = 503.50 KB/sec +++
ttcp-r: 2306 I/O calls, msec/call = 14.45, calls/sec = 70.87
ttcp-r: 0:32real
UDP: Amiga->PC
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  udp  -> 10.95.1.73
ttcp-t: 16777216 bytes in 28.34 real seconds = 578.12 KB/sec +++
ttcp-t: 2054 I/O calls, msec/call = 14.13, calls/sec = 72.48
ttcp-t: 0:28real
TCP: Amiga->Mac
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> 10.95.1.10
ttcp-t: 16777216 bytes in 41.66 real seconds = 393.28 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 20.83, calls/sec = 49.16
ttcp-t: 0:41real
TCP: Mac->Amiga
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
ttcp-r: 16777216 bytes in 30.18 real seconds = 542.88 KB/sec +++
ttcp-r: 2307 I/O calls, msec/call = 13.40, calls/sec = 76.44
ttcp-r: 0:30real
UDP: Amiga->Mac
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  udp  -> 10.95.1.10
ttcp-t: 16777216 bytes in 30.06 real seconds = 545.04 KB/sec +++
ttcp-t: 2054 I/O calls, msec/call = 14.99, calls/sec = 68.33
ttcp-t: 0:30real

(edit): I would say that these results are really good, especially considering that the Amiga is only powered by a 030@25MHz.

Have fun!


/Patrik

[patrik]

@Ilwrath:

Actually, curiousity got the best of me, so I ran all relevant tests. Here's the results:

Cheers :=)

PC -> Amiga (UDP)
* Clearly invalid results [>11000KB/sec -- not sure why?]

As UDP does neither guarantee or check if data arrives, the pc will push as much packets it can at the current link speed, disregarding if the other end can cope with it or not, which is the reason for the high results. In this case, it will just end up with the switch throwing most packets away and the ttcp app on the Amiga shouldnt consider it as a completed session as it hasnt been able to recieve enough packets. I also tried a this with a crossover cable to get the linkspeed down to 10mbit, but as the Amiga I benchmarked cant cope with a fully utilized 10mbit connection either, that didnt work. In short - for the UDP tests of ttcp to work, the reciever must be able to cope with the packet-rate the sender can manage to send.

Just as a sidenote, the results you got with your A4000 + CSPPC + A2065 look very similar to the results I got with my A4000 + CSPPC + A2065 ;=).


/Patrik

[patrik]

@Cass:

I use a 10BaseT transceiver connected to the AUI-port of the A2065 to get a RJ45 connector. There are lots of different transceivers, I happen to use this one as I got it with the A2065 card.

It would indeed be possible to build a ZorroIII 10/100Mbps ethernet NIC. The bandwidth of the ZorroIII bus would be adequate if the NIC was constructed well as the ZorroIII bus can do over 15MByte/Sec with a CV64 and an appropriate CPU-card (such as for example the Commodore original 030 card or the CSPPC). To fully utilize a full duplex 100mbit connection, a bandwidth of approximately 25MByte/sec would be required, but this bandwidth would seldom be needed during normal operation.

Anyhow, the ZorroIII bus would not be the limiting factor of speed if such a card would be produced. Given that the card was constructed in a really good way with support for scatter/gather dma transfers, delayed interrupts, hardware checksum offload and maybe even more features to reduce system load, there would still exist some big performance hinders.

These are a combination of the SanaII-standard, the current TCP/IP stacks and the socket api (the socket api is almost impossible to keep given the current AmigaOS functionality if one wants to remove the negative impact it has on performance). As it is right now, these hinders put too much stress is put on the memory-system and cpu... and the fact that almost none of the existing cards support any of the hardware features I listed above (the one on the DKB-wildfire supports DMA transfers if I am not mistaken) makes the situation even worse... ever wondered why Amigas even when equipped with a 68060 currently has problems fully utilizing even a 10Mbit link?

If anyone is interested in this subject, a good article can be found  here.


/Patrik

[patrik]

@All:

I guess most of you already know this, but to get the most speed out of your network-cards, use AmiTCP/Genesis as it is the most efficient TCP/IP stack available... if Olaf Barthel by any chance would be reading this: I would gladly buy a copy of your Roadshow TCP/IP stack right now .

Have fun!


/Patrik

[patrik]

@StormLord:

As far as I know, it doesnt support VPN. If you want to use AmiTCP/Genesis with a VPN connection, my advice would be to get a old pentium, run Linux on it and use it as a VPN router. This solution should be much faster as the Amiga will get a faster TCP/IP stack plus it is relieved from the VPN job which then is done by the Linux box instead.


/Patrik