Does AROS have the same problem with slow internet connections?

> there are two distinct things, and one of them is [...] a side-effect of a dated TCP/IP
> stack. [...] about 350-400kB/s from WAN: [...] it has to do with the TCP window
> scaling feature, that is not implemented in MorphOS TCP/IP stack [...]. Without
> this feature, the speed is actually quite dependant from the roundtrip time between
> the server and the client, which explains why it's faster if the server is close.

Will the updated NetStack that is based on newer FreeBSD code and is supposed to ship with MorphOS 3.2 remedy or mitigate this issue?

Yes, the new stack supports window scaling and it definitely improves the speed / stability of the connection (in some cases dramatically so).

There might still be scope for improvements to the sungem device, but that's not my field

> Yes, the new stack supports window scaling and it definitely improves
> the speed / stability of the connection (in some cases dramatically so).

Fantastic news. Thanks :-)

Nadir,
Quote:

Will there also be support for fragmented packets?

You mean datagram fragmentation? Even the old stack should do that. One issue that is still outstanding relates to SACK (selective acknowldgement of received packets). This is a TCP level feature which is very useful when packets arrive in the wrong order.

I'm not sure if this will be included in the 3.2 release. I currently work on two different FreeBSD trees: 2.2 (with some patches from latest repository) and 4.11 (+patches). It might be that only the former will be ready for the next MorphOS release and it does not have the SACK feature.

There are big changes in the 4.x tree and it will take some more time to make sure everything works so 4.x (or an even later release) might have to wait unfortunately. Good news is that compared to the old stack, you will see a large improvement anyhow. Even a FreeBSD 2.2 based stack is newer than anything else on our systems.

/Nicholai

Experiencing same TCP problems as everyone else using OWB on a Powerbook.
Seems like a LOT of packet loss.
Causing web pages to take a long time to finish.
Downloads are all over the board, usually dropping to 20kbs from 200kbs for no reason.

Since the upgrade to 3.1, all issues with OWB 1.18 have vanished! The net is fast and friendy! Ty FAB

Nadir,
Quote:

I bow to you, Sir!

Quote:

There seem to be a few different types of network 'problem' so it's probably more useful to describe your issues rather than pointing at some nebulous thead and saying 'me too'.

- Not everyone experiences problems
- Many people are describing problems on the Mac-Mini rather than the Powerbook.
- Problems are not just related to using OWB, but any type of network transfer (FTP, HTTP, SMB)

Anyhow, Network Statistics window tells you how many packets have been dropped. (Enable Netlamps screenbar module and double-click on it)
Severe packet loss due to the stack (rather than crappy cabling/router) would be a new issue afaik.

EDIT:

Quote:

Try testing with a fixed download size of 12Mb download size at testmy.net. I've found it a more reliable way of measuring Cx speed than just downloading random files and guestimating. Also, if you create a profile you can moitor your network speeds over time.

[ Edited by boot_wb 04.01.2013 - 14:16 ]

These kind of swings in download speed is probably due to the fixed window size in the old stack. I also used to have this kind of behaviour.

> the TCP window scaling feature [...] is not implemented in [...] OS4

According to olsen, Roadshow for OS3 has this feature:

http://www.amiga.org/forums/showpost.php?p=721301

I don't know if that's also true for the version OS4 comes with, though.

OWB crashing all the time.
Fresh 3.1 install and nothing but latest OWB.
Pages will suddenly halt and die and someitmes OWB says 'application meditating'.
Downloads dropping like crazy.
Took this laptop to work and downloads worked fine but OWB still crashing on loading web pages.
Takes just a few minutes to start crashing,
tigerdirect.com - select SSD catagory, page wil always crash after loading half the page.


System:
Best Powerbook Apple ever made.
Nuf said.

To be honest, I've always found the issue of the sungem stalls and varying speed pretty swept under the carpet, either denial it exists any more or "well we can't replicate it" and then being walked through the usual (blame the router, blame your ethernet cable, dns settings etc.). In the end after having everything from my router, ethernet switch, wiring etc. blamed (all of which work fine on other devices in the same room), I got a Belkin USB to Ethernet adapter, works great with that so I disabled the internal etherner on my mac mini.

Agree, it would be nice to finally admit this is an issue and to try and solve it at last, as it would free up my second usb port again.

[ Edited by danwood 06.01.2013 - 21:31 ]

hooligan,
Quote:

I think this is the Sungem issue, it happens with me too using internal ethernet. On a USB to LAN adapter it does not, it loads smoothly every time, although ocassionally I get the system locking up on me randomly :/

@amigadave

Did you try the new Roadshow 68k stack ?

with the current netstack in morphos OWB crawled and file transfers
went from between 500KB <-> 6-7MB in jumps up and down

this new(old) tcp/ip stack, even as emulated 68k code have boosted the 2 morphos machines I have to be the fastest in my lan (but in ramdisk)

I think i sat an hour, just copying files back and fourth....seing 35-36MB/s filetransfers on a Morphos machine.

Ofcource reality hits when you want to do real work, as samba i have is slow and 68k, and harddisk speed and ide.device takes its cpu, so in the end, writing to a SFS SSD the Powerbook ends up on flat 6-7-8MB/s and no jumps/jitter in the process, but cpu load was pretty high.

A ppc port of samba would help..

Quote:

Check Aminet (there's also an update as well as the main archive),

Although I prefer using Apache and wget myself, since I dound that the samba daemon throttles the cpu on my mac-mini after a while.

[ Edited by boot_wb 15.01.2013 - 10:31 ]

accessing your files with wget would give less cpuload but accessing your files from remote or linux/win/phone locally you cant with wget.

Watching movies with Mplayer streaming from my smbfs mounted nas or win7 directly from Ambient is quicker that to wget your movie first to harddisk.

But smbd works better than smbfs in terms of transferate, i.e
copying a file from MorphOS: as a smb share, but you copy from win7 gives
pretty good transferate(14-15MB/s), but if you smbfs mount your same win7 harddisk and copy the same file from MorphOS: from Ambient windows gives me 1-2MB/s :(

It might be totally wrong, but Roadshow 68K speeds seems to increase over time, as if Trance JIT works faster over time.....after a reboot, transferates are lower than if its been running a while...?

> as if Trance JIT works faster over time

Trance JIT doesn't work faster over time. It's just that it kicks in only after a code segment has been executed by the slower interpretative emulation a certain number of times. This is done that way as for code segments which are only run once (in a while), the interpretative emulation is actually faster than Trance JIT due to the compile overhead of the latter.
It may be that this is what you're experiencing regarding Roadshow speed.

> One issue that is still outstanding relates to SACK (selective acknowldgement
> of received packets). This is a TCP level feature which is very useful when
> packets arrive in the wrong order. I'm not sure if this will be included in the 3.2
> release. I currently work on two different FreeBSD trees: 2.2 (with some patches
> from latest repository) and 4.11 (+patches). It might be that only the former will be
> ready for the next MorphOS release and it does not have the SACK feature. There
> are big changes in the 4.x tree and it will take some more time to make sure
> everything works so 4.x (or an even later release) might have to wait unfortunately.

The MorphOS 3.2 release notes reveal that dynamic window scaling and Berkeley packet filter support have been implemented, which is really nice. Would you please tell if you went for the BSD 2.2 or the BSD 4.11 version?

Thought I would share.. problems with low download speeds on my Mac are gone, I get 3.6-3.7MB/s instead of crawling I used to get. Ok so its nothing to be proud of as it should be over twice better but still a heck of an improvement.

Quote:

If its possible to tweak internal tcp/ip stack settings in new Netstack as we can in Roadshow 68k, you could probably get most of your missing half, but we need documentation if its possible with env-vars or adding values to configfile from Morphos devs...

Quote:

It is based on FreeBSD 2.2 but with a few additions from FreeBSD8/9.

Btw, one feature that didn't make it into the release version is SACK support which I finalised only one day before the internal feature freeze and we decided to test it more before including it in a user release. It has a very positive impact on networks with a lot of packet drops / delay so that packets acks arrive out of order (which seems to be the case in my home nw for whatever reason...). On some networks it doesn't help at all basically because there is no problem in the first place

I am still making progress on the newer codebase, so let's see what happens with that as well

Quote:

> It is possible to control these kind of settings through the ARexx interface or through
> env:sys/net/NetStack.config. [...] Future releases will however support a larger set of
> variables that can be controlled.

Now that NetStack settings can be controlled via Network prefs GUI in MorphOS 3.4, is NetStack.config still used or can this file be deleted?