Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

A1200 IceDrake Bad Network Performancepage  1 2 

Jesus Reyes

Posts 4
07 Aug 2022 15:35


Hey there,

I am using a Icedrake with OS 3.2.1, all really neat and no patches added, just the one from the SOLO installation. All works more or less great except the network. Even on "local network" I only can get around 4 to 8Kb of speed when doing just simple FTP transfers from a NAS that I have at home to the A1200.

Why such a bad speed? Is it there any bug or fix I can apply to make the Vampire's network faster?


Gunnar von Boehn
(Apollo Team Member)
Posts 6194
08 Aug 2022 07:08


Network Speed of around 1 MB/sec is normal.




Andy Hearn

Posts 374
08 Aug 2022 09:14


whats your install setup like? where did you get the driver stack for the Icedrake devicecs?
 
  install the latest SAGA drivers.
  i found i had massive network issues with my icedrake initially, but it was down to my software setup.
  it was tied to an odd version of the V4 ethernet device that was installed in my coffin distribution. i can't remember what version i was having issues with, either coffin58 or 59 - from when i switched over from my V1200v2 to Icedrake,
  someone helpfully pointed me over to *i think* version3 of the v4's ethernet.device driver and it's been fine ever since.
 
  my icedrake 1200 is sitting wating to get plugged back in at the moment, as i'm having a sort out at home. but i'll set it up tomorrow and confirm versions of the v4's device driver
 
  but as a starting point, and if you're running a "stock" amiga OS, and not something specially tuned for the vampire, install those SAGA driver packs


Jesus Reyes

Posts 4
08 Aug 2022 09:50


I installed using ApolloBoot SOLO R8, so I guess it uses the latest SAGA drives and same for network. I will check the versions tonight when back home and post it here. I never tried to upgrade any library or device since I used an "official" way of installing the OS. By the way my setup is a stock 1200 from Commodore with a Icedrake v4, no extra packages or anything on the OS apart from what ApolloBoot SOLO R8 does when installing the AmigaOS.


A. Dewart
(Needs Verification)
Posts 12/ 1
08 Aug 2022 10:00


Gunnar von Boehn wrote:

Network Speed of around 1 MB/sec is normal.

1mb/s should be considered normal ?
could you explain why ethernet port cannot reach higher throughput?
is it supposed to improve later on, or is this some kind of hardware limit ?



Andy Hearn

Posts 374
08 Aug 2022 10:43


I could get about 17-18megabits a second out a PCMCIA card, and what i suspected there is a limit of the gayle chip handling the PCMCIA slot - is going to have the same sort of limits as the standard amiga IDE port.

the latest SAGA drivers got me up to about the same - about 12-14megabits to where i felt i could remove my netgear FA411 from regular service

i can't remember exact numbers but it was close enough.
So no, there isn't a hard stop at one megabyte a second, but it's a good "line in the sand" number for what we should be aiming for. :)


A. Dewart
(Needs Verification)
Posts 12/ 1
08 Aug 2022 12:16


Andy Hearn wrote:

I could get about 17-18megabits a second out a PCMCIA card, and what i suspected there is a limit of the gayle chip handling the PCMCIA slot - is going to have the same sort of limits as the standard amiga IDE port.
 
  the latest SAGA drivers got me up to about the same - about 12-14megabits to where i felt i could remove my netgear FA411 from regular service
 
  i can't remember exact numbers but it was close enough.
  So no, there isn't a hard stop at one megabyte a second, but it's a good "line in the sand" number for what we should be aiming for. :)

onboard ide does 10+ megabYte per second.
i would have expected the ethernet port to be at least as fast as that.

1mb/s (megabyte) seems quite slow.
is it a driver problem ?
on aros i don't think they have such limitation in throughput, so it's probably not a problem with the network stack itself.



Gunnar von Boehn
(Apollo Team Member)
Posts 6194
08 Aug 2022 12:47


a. dewart wrote:

onboard ide does 10+ megabYte per second.

 
The SAGA FASTIDE Controller reaches up to 20 MB/sec.
Its by far the fastest IDE for Amiga.
And our IDE driver was written by myself, so no wonder that its lightning fast ;-)
 
a. dewart wrote:

i would have expected the ethernet port to be at least as fast as that.

 
Of course Ethernet has a lot more protocol and driver overhead than IDE.  On ApolloOS I do reach network speed of about 1 MB/sec to the internet. In Local Lan connection I can reach more.

The ethernet hardware can physically do nearly 10 MB/sec.
Internet delays and network stack will limit this.
The results will vary depending on the used stack version.



A. Dewart
(Needs Verification)
Posts 12/ 1
08 Aug 2022 15:11


@biggun
  you didn't say if that is a problem with driver or network stack ?
  and, is it planned to improve the situation to match a bit better with hardware specs ?

as you said hardware specs allows for 10mb/s
current reality over internet is 1mb/s
and current reality over local net is "more" (how much is "more" ?)
so, right now the situation is 1/10

i would expect situation to improve at least to 5/10
..to be in the fair range.


Roy Gillotti

Posts 517
08 Aug 2022 16:45


For AmigaOS I have the best speeds with Roadshow vs Miami and others. It's still not as fast as PCMCIA ethernet in the case of my V2 A600 setup. But the V2 uses cheap off the shelf ENC28J60 modules and the limitations of the SPI bus over cheap .1mm headers.


A. Dewart
(Needs Verification)
Posts 12/ 1
08 Aug 2022 18:58


@roy
but biggun said it could go up to 10mb/s.
that is why i suspected the bottleneck would be software based (driver or network stack).
what you're saying would, on the other hand, suggest hardware limitations, yes ?
it would be nice to have clear answer on that topic.
whether a future software update would improve things at least 5x times or if faster net speed would require a new board with different components/different design.

i would have similar questions on the usb ports too, but that would be for another topic.


Gunnar von Boehn
(Apollo Team Member)
Posts 6194
08 Aug 2022 19:22


a. dewart wrote:

  @roy
  but biggun said it could go up to 10mb/s.
  that is why i suspected the bottleneck would be software based (driver or network stack).
 

 

I measure 1 MegaByte/sec Internet speed and you speak about "bootleneck" -
Where do you know if my Internet connection can even do more?
 

a. dewart wrote:
 
  on aros i don't think they have such limitation in throughput, so it's probably not a problem with the network stack itself.
 

Where do you know this?
Can you explain us in which information your claim is based?


A. Dewart
(Needs Verification)
Posts 12/ 1
08 Aug 2022 20:20


Gunnar von Boehn wrote:
 
    I measure 1 MegaByte/sec Internet speed and you speak about "bootleneck" -
    Where do you know if my Internet connection can even do more?
 
    Where do you know this?
    Can you explain us in which information your claim is based?
 

 
- i do _not_ know if your internet connection can do more.
    (i do not need to know that, to ask my question).
- i do _not_ know if aros network stack is crippling the throughput.
    (i do not need to know that, to ask my question).
 
but,
- i _do_ know you said hardware can physically do 10mb/s.
- i _do_ know you said 1mb/s is to be considered normal.
 
so my "claim" is simply that:
a product that can only reach speed of 1/10th of target specs is probably going through a bottleneck somewhere.
 
so i simply asked what was the bottleneck ?
and, would this improve in the future ?
this is a simple question.
you could say: it is aros network stack fault.
or you could say: it is driver fault.
or again: it will improve in the future. or not..
i do not know, that is why i'm asking.


Gunnar von Boehn
(Apollo Team Member)
Posts 6194
09 Aug 2022 08:22


Lets prevent some confusion here:

Mind that megabit and MegaByte are not the same.
A Byte has 8bit.
 
MB != mb
 
MB means Mega-Byte
mb means Mega-Bit
 
1 Mega-Byte = 8 Mega-Bit
 

Lets not confuse different hardware.
On this thread people spoke about both V2 and V4 hardware:
Mind that the V2 and V4 hardware are different.
 
Some V2 have to rely on using PCMCIA network cards.
 
Some V2 have an expansion port where you can connect a 100 Mbit ethernet module. This module can in theory reach 100 Mbit but the expansion port can not reach 100 Bit. So you can reach 25Mbit at best. But this is still good enough for perfect internet surfing. 
 
The V4 has an SAGA-ethernet chip integrated in the Amiga chipset.
Like real Amiga chips it can do on its own full 64bit DMA (direct memory access). This SAGA ethernet chip can do full 100MBit downloads without needing any CPU power. This is fastest Ethernet chip ever available for Amiga.
 
The V4 Ethernet chip also has some extra feature like checksum off-loading.
 
Ethernet protocols calculate checksums.
The SAGA chip can do this own its own - without needing the CPU. But currently no Amiga software stack does supports this. Ethernet stacks are sending packets and working in layers, and you create packets around again another packets are wrapped. I my experience the existing software stacks for Amiga are all not optimal. Also the AROS stack is by far not optimal. Its internal design adds a lot overhead and delays.
 
 
Wanting to reach 100MBit with the current Amiga stacks is totally impossible.
 
But do we really need to have 100Mbit (=) 10 MByte/sec?
I personally don't think so.

1MB/sec internet speed means that a whole Amiga floppy can be transferred in one second. In local network connected this will be higher. But I think this speed is very usable. We did reach in local test lot higher speed, like 40 Mbit. Rework the AMiga network stacks would be cool and reaching a lot higher would be nice for benchmarks. But with current speed I can live. Its useable enough.

What do you think?
 


A. Dewart
(Needs Verification)
Posts 12/ 1
09 Aug 2022 11:01


you're right, let's clear some details from my end too:
- i always speak in effective megabyte per sec.
- i speak of v4sa as test machine.
(because v4sa is supposed to showcase apollo/saga arch the best).

now, from your answer i understand you're saying the bottleneck are the amiga tcp/ip stacks. you seem to indicate that especialy aros network stack is not optimal. yet, i see that aros (68k distro "aros one") do reach 2.4mb/s over internet.

here is such an example:
EXTERNAL LINK 
let's talk about this case.
downloading from youtube is not the best way for testing throughput it's dependent on a few factors, like the route between youtube and the user, of course. but also youtube might cap the speed to share the overall load on its platform for everyone to access it. so using youtube is not best to test bandwidth, that's clear. still, this "real life" case shows download at double the speed you indicate as typical peak bandwidth on apollo/saga platform.

i also saw the same test case on morphos where the download speed was 3.5mb/s. for example here:
EXTERNAL LINK 
what i think ?
i think 1mb/s is 1/10th of hardware specs. i understand reaching 10/10th is not reasonable and there's always some overhead and speed loss. still, having for goal 8/10th and actually reaching 6 or 7/10th would look like well optimised and nice situation.
what we have here, 1/10th, look like seriously crippled throughputs.
i would suspect the driver too. since the links i've shown above demonstrate aros tcp/ip stack (on 68k) already more than double the speed you quote as typical/normal/peak performance in current situation. and as for reaching 5mb/s (40mbit/s) on local network, i'll believe it when i'll see it. for now, networking capabilities of apollo/saga platform seems kind of underperforming to me. based on what i see.




Robo Kupka

Posts 50
09 Aug 2022 14:33


a. dewart wrote:

 
  here is such an example:
  EXTERNAL LINK 
  let's talk about this case.
  downloading from youtube is not the best way for testing throughput it's dependent on a few factors, like the route between youtube and the user, of course. but also youtube might cap the speed to share the overall load on its platform for everyone to access it. so using youtube is not best to test bandwidth, that's clear. still, this "real life" case shows download at double the speed you indicate as typical peak bandwidth on apollo/saga platform.

What is "real life" case from that video ? That is WinUAE... unknown emulation speed, unknown network driver, unknown underlying PC hardware.

Gunnar said, that the HW implementation of Ethernet on V4 has some checksum offloading implemented. I assume that he was speaking about Ethernet frame checksum calculation (that´s layer 2 of OSI model). Everything else related to IP protocol (layer 3) and upper levels - TCP/UDP (layer 4) is handled at TCP/IP software stack level.
As Gunnar said, currently available network stacks for Amiga do not use even the hardware Ethernet frame checksum calcucation, so even that is calculated in software.
Thus I conclude, that 150 MIPS CPU is not capable to handle network traffic of 100 Mb/s, even thgough DMA could process 100 Mbps stream easily.
Even on a PC hardware, a typical ethernet frame has 1492 bytes of payload. Network hardware receives such frame, calculates its checksum and if it is OK, accepts it into buffer, then calls INTERRUPT to CPU, which then Network driver in the OS processes the frame, then TCP/IP stack.... simply, lot of CPU intensive work.
If new revisions of Amiga's TCP/IP stack utilize 64bit DMA Network hardware properly,  I would guess that 10 - 20 Mb/s (not 10-20 MB/s !) is realistic goal for V4. For higher speeds, simply there isn't enough HW offloading, nor enough CPU power.

If I recall, first RASPBERRY PI with 1-core ARM CPU@700 MHz had 100Mb/s ethernet interface, but was capable to process barely 35-40 Mb/s, event though the network stack was much more mature.


A. Dewart
(Needs Verification)
Posts 12/ 1
09 Aug 2022 15:24


@robo
  what you say contradicts what gunnar said.
  he already said that localy he can reach 5mb/s.
  (that's 40mbit/s but you're saying it can only do 20mbit/s)
 
  .. also i dislike speaking in term of mbit/s,
  i prefer to always talk about effective speed.
  so gunnar said he can do 5mb/s, and you say v4sa cannot do better than 2.5mb/s. that is why i'm saying you're contradicting gunnar.
 
  so if he can reach 5mb/s on local network that mean he can also reach very similar speed over internet (if his provider gave him good internet access, and if net route is good between him and some online site). i'd still like to see a video showcasing this, but let's give him some trust, after all he's the engineer behind this project. if he say it works at 5mb/s then fine. i hope someone will post a video showcasing it. both local and online hopefully.
 


Robo Kupka

Posts 50
09 Aug 2022 16:49


@A.Dewart
  sure, I realize that it might be confusing. When someone says 5mb/s, I assume it means 5 megabits/s and it is common practice in IT to write "Mb/s" as megabits/s, and MB/s as megabytes/s.
 
  If gunnar says that 5 mb/s was truly meant as 5 MegaBytes/s, which equals to 40 megabits/s, that is a fantastic result for V4 hardware and current TCP/IP stack on 160 MIPS CPU.
  But I would be happy if it was even less, as I said, 25 Mb/s would be still great.
 


A. Dewart
(Needs Verification)
Posts 12/ 1
09 Aug 2022 17:22


@robo
  you cannot read his post above ?
  he is clear about it. he says localy he reaches 40mbit/s (which is 5mb/s).
   
  also, you're introducing a lot of confusion because you keep switching between mbit and mb (megabyte) in your posts.
  if you want to talk about megabit then write it as mbit, as i do.
  that way there is less confusion.
  for myself, i calculate disk space and file sizes in bytes, so when i talk about network speed i equally talk in bytes per sec.
  the only times i'm forced to think about mbit is when i look at a network card box in a shop because it's advertizing in mbit/s.
  reality is always about mb/s (megabytes per sec).
  when you download a file, you get told how much of the file you have downloaded so far, so speed indication is in byte per sec, because file size is also calculated in bytes too.
   
  anyway, my v4sa reaches 1.6mb/s (effective megabytes/s).
  and when i do that, the cpu usage isn't maxxed out. not even 50%.
  so there's plently of resources left. it shows around 30mips (using flype's cpu monitor).
  so i guess the bottleneck is primarily the driver, then in 2nd position network stack. for sure there's room for improvement. and i hope future updates will bring that. for comparison, exact same system on fsuae, configured to generate 160mips 68k, handles 15mb/s without breaking a sweat. so the cpu crunch power is not the limiting factor. when i burn that exact same system on cf card and use it on v4sa, i get only 1.6mb/s speed (that is a x10 downgrade). to give a bit of context: i'm using optic fiber line, linux does 30+mb/s here.
   


Steve Ferrell

Posts 424
09 Aug 2022 18:14


a. dewart wrote:

@robo
  you cannot read his post above ?
  he is clear about it. he says localy he reaches 40mbit/s (which is 5mb/s).
     
 

It's tough to follow anything you're saying because you seem to have totally missed the point that when the industry or IT folks talk about megabytes and megabits, the abbreviations are CASE-SENSITIVE. 

MB = megabytes, Mb = megabits. 

B = bytes, b = bits
M = mega, m = shouldn't really be used.

After reading the sentence you wrote in the quote above, I immediately stopped reading any further because nothing makes sense due to the lack of case sensitivity.


posts 30page  1 2