Welcome, Guest. Please login or register.

Author Topic: IBrowse2.3 download transfer rate.  (Read 6345 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline patrik

Re: IBrowse2.3 download transfer rate.
« on: August 13, 2004, 06:07:59 PM »
@AmiWalker:

IBrowse writes data to files in a very inefficient way. This often results in comical transferrates even when transferring data from a computer on the same local network.

There is one thing you can do to lessen the impact of IBrowses inefficient file-writing and that is using a better filesystem than FFS. Anyhow, first of all you should investigate wether it is IBrowses inefficient file-writing that is causing this by downloading to ram: instead and see if you get any better transferrates.

If you do get better transferrates when downloading to ram: I would recommend you to start using the SFS filesystem on the partition you are downloading stuff to and be sure to allocate it say 128 Buffers. I would actually recommend you to use SFS on all your partitions and harddrives as it is much faster than FFS.

To measure the raw tcp and udp speed the TCP/IP-stack plus NIC combo can manage (Genesis plus X-Surf2 in your case),  you can use ttcp and measure the speed between your Amiga and PC.
Amiga version of ttcp
Windows version of ttcp
Sourcecode for ttcp

Miami is generally much slower than AmiTCP and Genesis (which is practically AmiTCP with a GUI ontop to control and configure it) so you wouldnt gain by switching to it. Though if you have doubts about this, use ttcp to measure which one of the TCP/IP-stacks that are the fastest.

A boot-rom would not help the speeed in any way. Its sole purpose is to contain code used to boot from a NIC. Also, I am quite positive that there doesnt exist any boot-rom for any of the Amiga-NICs which has a boot-rom socket.


(edit):

I get 250KByte/Sec when downloading with IBrowse from a webserver on the same local net to a partition with SFS. If I instead download to ram: I get 300KByte/Sec. My A4000 has a A3640 cpu-card, a A2065 network card and I am using AmiTCP 4.3. As my A4000 is quite a bit slower than your A3000, you should be able to get transferrates atleast as high as those when downloading from a source capable of supplying data at those rates.


/Patrik
 

Offline patrik

Re: IBrowse2.3 download transfer rate.
« Reply #1 on: August 13, 2004, 07:47:46 PM »
@blobrana:

I have to agree with you that the combination FFS and IBrowse doesnt cause bad performance when downloading. I created a new partition to try it and got quite the same results.. doesnt it feel just great when you realize that you are totally wrong? ;=)

Though I still have to say that IBrowse seems to write very little data to the harddrive per call to the filesystem and that is just not very efficient. Just watch the harddrive-led when you are downloading from a webserver with a good connection on say your local network. I can almost hear the led scream: "lots of very small requests, lots of very small requests, lots of very small requests - iiiih"..;) in other words, what I am trying to say is that there is way more activity than there should be at such transferrate. Then try copying a file from a slow device like a CD-ROM set to 2X or 4X which has a similar transferrate to what you get when downloading something from a local net and watch the harddrive-led - it isnt screaming anymore :D.

If any IBrowse-programmers are reading this forum I would like to tell them to buffer more data before writing it out to disk and preferably give one the option to set the buffersize in the preferences.

Btw, he shouldnt have to worry about PPPoE as he has a router connected to the Cable-Modem which should handle that part for him.


@Amiwalker:

Just thought of something - you dont happen to have set the X-Surf2 to work in full-duplex mode? If so, you must change it back to half-duplex mode. This is as the X-Surf2 and most 10MBit cards isnt capable of auto-negotiating for linkspeed and duplex-mode with the other end of the line (your router). When your router senses that the X-Surf2 doesnt want to autonegotiate with it, it will assume that the X-Surf2 is working in half-duplex mode per default even if it is actually working in full-duplex mode. If that is the case, it will render _TONS!_ of collisions on the networking, slowing the transferrate to a crawl.

In the file ENVARC:Sana2/x-surf.config, set the variable 'Unit0.FullDuplex' to '0' and then reboot to make the X-Surf2 work in half-duplex mode.


/Patrik
 

Offline patrik

Re: IBrowse2.3 download transfer rate.
« Reply #2 on: August 14, 2004, 03:08:33 PM »
@amiwalker:

Hmm... not that it should make any difference, but try moving the ENVARC:Sana2/x-surf.config file to say your WORK: or similar partition and then reboot and see if it makes any difference.

You should measure the raw tcp/udp transferrates between your Amiga and pc by using ttcp to see if the Amiga is working properly on with the switch part of your router (I assume your router is one of those with one WAN-port and about 4 10/100 switched ports).


/Patrik
 

Offline patrik

Re: IBrowse2.3 download transfer rate.
« Reply #3 on: August 14, 2004, 03:50:54 PM »
@amiwalker:

---- Quick guide to how to use ttcp ----

- I will use amigaip instead of your Amigas ip-address and pcip instead of your pcs ip-address.

- I assume that you have installed ttcp in your path and that the executable for it is named 'ttcp'.

- You shall run ttcp from a shell/cli/command-prompt on both the Amiga and pc.

- I will call the shell/cli/command-prompt on both the Amiga and pc shell.

Below I will describe what to input in the shell and the order of it on the Amiga and pc.

To test the speed when sending tcp-packets from your Amiga to your pc:
1. PC: 'ttcp -s -r'
2. Amiga 'ttcp -s -t pcip'

To test the speed when sending tcp-packets from your pc to your Amiga:
1. Amiga: 'ttcp -s -r'
2. PC: 'ttcp -s -t amigaip'

To test the speed when sending udp-packets from your Amiga to your pc:
1. PC: 'ttcp -u -s -r'
2. Amiga 'ttcp -u -s -t pcip'

To test the speed when sending udp-packets from your pc to your Amiga. How to make this test is only included for reference. It will most likely not work as the protocol used for udp just sends packets as fast as possible without checking if it was recieved correctly. This in combination with the Amiga not being able to recieve packets as fast as the computer in the other end is sending them which usually is linespeed (100MBit in your case?) makes this test fail. Usually the Amiga doesnt even understand the test is done and the pc reports speeds of over 10MByte/Sec.
1. Amiga: 'ttcp -u -s -r'
2. PC: 'ttcp -u -s -t amigaip'


/Patrik