ACA500+ ACA1234 X-SURF500+ True network speed

Caution: Non registered users only see threads and messages in the currently selected language, which is determined by their browser. Please create an account and log in to see all content by default. This is a limitation of the forum software.


Also users that are not logged in can not create new threads. This is a, unfortunately needed, counter measure against spam. Please create an account and log in to start new threads.

Don't Panic. Please wash hands.
  • I have a question about this set because I really don't know that this is the maximum that can be achieved.


    With a 220mbit/s network I can achieve a maximum of ~1.5mbit/s when tested in Os3.2 SpeedTest. Which should give a maximum of ~187kbyte/s. In fact, I'm getting around ~36kbytes/s. How does this compare to the advertised 100mbit/s. Generally, I achieve ~300kbit/s on Plimpbox and realistically about ~18kbyte/s. What is going on ?

  • I'm anticipating the question, the bandwidth of the line is 50 Mbit/s, i.e. about ~6 MBytes/s, and this is a guaranteed speed. I can easily get 5MBytes/s from a torrent on a PC/Mac from the same network.

  • Raw link speed is never the performance you get for file transfer.


    Network has multiple layers that all need CPU attention. Network is particularly CPU-heavy, as packet headers need to be copied back and forth for handing over between the different software layers. This is actually something we've worked on with AmiTCP, as we have managed to eliminate one of these copy-steps, so our networking cards operate faster than others, given that you compare with the same CPU speed.


    If the networking part has been manages (where you lose most of the performance due to said copy operations, but also to checksum checking), you continue losing on the next layer, the filesystem protocol and handshaking - for a fast internet line, it will be comparable to making a connection to a local NAS, as handshaking/protocol overhead will kill most of the performance.


    For a proper assessment of your setup, you should clearly specify what protocols you are using. Modern protocols are mostly slower than old/simple ones. If you truly want to max out what the networking card and TCP/IP stack can do, you should stick to good old FTP. If you want to access files remotely using CIFS, expect major slowdowns for reasons axplained above.


    Jens

  • This explains exactly my surprise! Thanks. I have a question whether using a different stack, e.g. RoadShow, will be more efficient than AmiTCP4. I checked MiamiDX but it is losing too many packets. Reagrds

  • Roadshow is really fast, and may come close to what we achieve with AmiTCP and our optimization of eliminating copy-steps. Think of it like a "side-entrance" into AmiTCP that we have created in order to minimize overhead that is created with the sana2 driver protocol.


    It's worth a try, but I don't expect Roadshow to me much faster (if it's faster at all). Roadshow is a very good and efficient stack, but I'm hesitating to recommend it, as you already have a very good stack for free with your X-Surf-500.


    Jens

  • TCP is a rather high-level protocol, and with a HW interface as fast as the X-Surf-500, performance is very CPU bound.

    True (not artificial, not hypothetical) network speed can be measured for example with the program ttcp. Of course you would run the other end on a very fast computer such as a Linux PC.


    68000, 7MHz: 139,53 KB/sec

    68030, 50MHz: 1770,66 KB/sec


    In my opinion that's pretty cool, and our adaptations to AmITCP make it not only fast, but also very user-friendly. So there's probably no need to change to Roadshow, since our solution is at least as good. But Roadshow is also a very decent product, and if you want to support the Roadshow author, you can do this, and even continue to use our Network GUI with Roadshow.


    Also a very useful test is to measure the time needed to copy for example a huge file over NFS or SMB from/to a Linux server.


    Edit: For comparison, and to show how well our hw/driver/TCP stack combo performs:


    68060, 50MHz: 1561,78 KB/sec


    That was on an Amiga 3000 with Cyberstorm MK-II 68060, X.Surf-100, and Roadshow. IIRC with AmiTCP that should be somewhere between 1900 and 2000KB/sec.