Re: Embedded Ethernet

Since the ATMega128 can't handle100mbps (ISA bus is too slow) isn't the issue is moot?

one

something

"Brett" wrote in news:bdcbjd$r4920$ snipped-for-privacy@ID-184277.news.dfncis.de:

Line rate and the CPU/bus's ability to handle this speed are not intertwined. If the 100Mbit controller can buffer a full frame at 100Mbs then it gives the CPU the ability to accept 100Mbs frames. Whether the CPU and bus can keep up with large amounts of these frames is another issue. I suspect that this design requires high line rate speed but not high data rates. Kind of like running a serial at 115kBaud but only sending a couple of characters per "long" unit of time.

Craig

The RTL8019 AS is 5 V only making it not the ideal candidate if you have a 3.3V CPU. Another option is the Cyrrus logic cs8900a which is also just 10Mb but available in a 3.3V variant. There is a 100MB chip with ISA interface available from SMSC. The partnumber is something like 9xxxx. You should easily find it if you browse their pages.

HTH

Markus

For 10BASE-T, the SMSC LAN91C96 is a simpler and lower power alternative to LAN91C111.

Tauno Voipio tauno voipio @ iki fi

There is a device out there that embeds a complete 10/100 interface, including a '186 micro, all the stacks etc into the Ethernet connector socket. Around AUS$75, probably around US$35. pricey, but simple, and very elegant.

the ad was in aaa local trade paper, I don't have that issue, but if you still need it I'll post it back. I think Circuit cellar ran an article using the same part in the last 2-3 months.

Al

Craig Rodgers wrote:

There is a 3.3V and 5V version of the MCU he's using.

I was looking at SMSC, however the bus looks like it's 16 bits data... is there an 8bit mode?

I would dare say they are, at least as far as dedicated point-to-point lines are considered. Having a high-speed capable line sit unused for the majority of the time is bound to be wasting some (costly) resource somewhere.

Shared-medium networks would be a different issue, obviously, but I don't quite see what could be the benefit of having that buffer to hold one complete Ethernet frame's worth of data on the embedded device's end of that line, instead of at the hub, switch or whatever is on the other end, where it'd usually be quite a bit easier to accomodate it.

The only thing that would really be improved by such a plan would be the latency, not the bandwidth. Sending shorter packets might be a better plan, in that case.

You forget the suits.

Being able to put "100 MBit/s capable" into the glossy feature sheet may be a sales advantage. Even though it doesn't make much sense technically...

at this stage I'd rather not switch cores simple because i don't have the development tools to play with the 8051 you've sugested, although it does look like a very attractive micro.

Do you have any idea who distributes the LAN91C111 in Australia?

Regards Craig

You have to build a byte cross-over buffer between the 8051 and 91C111 to bridge the bus-width gap. LAN91C111 does not support 8 bit bus.

Think twice if 100 Mbit/s is really needed - for a 8051 even 10 Mbit/s is an overkill.

Tauno Voipio tauno voipio @ iki fi

And how exactly will a point to point environment be improved by changing the ethernet from 100 Mbps to 10 Mbps? What costly resource does 100 Mbps Ethernet have that a 10 Mbps Ethernet does not have?

Yes, I can see that you don't understand. If you do all the math, you will find that data going to the embedded target can travel at the full rate and reach the target in less time. So the target can start working on the data sooner. Likewise data being sent by the embedded target will take less time if it travels at the higher rate freeing up the buffer more quickly.

Shorter packets may make some number smaller that is reported by software and you are equating with latency. But the real latency does not change, and in fact may get worse. No one cares how quickly a part of a buffer is sent. If the entire buffer needs to be sent before processing can begin, breaking it into smaller packets buys you nothing.

Projec Systems ETHERNET Controlled Web Server at:

-- Don McKenzie E-mail:

Add USB to your favorite Micro.

Atmel has a reference design kit.

They use Crystal CS8900 chip. (I am putting one in a new design too). The only drawback -- it is 10BaseT only, but I do not think you will use

Rudolf

one

something

Craig,

This is the chip you want. NE2000 like the RTL8019AS, but 10/100; 5v tolerant, but runs on 3.3v; LQFP-128 instead of TQFP-100. I've got one wired to an Mega128 in 16-bit mode, due to a similar migration from the RTL8019AS (for identical reasons). Only very slight extensions from your NE2000 driver to configure the 100Mb features.

For prototyping, this built board is actually cheaper than buying the chips from Singapore (at least for the USA), and he'll sell you 1-off chips too. Even with your proximity, you may find he offers a better deal. (ASIX wanted $60USD for qty 1, plus s&h from Singapore.)

The fellow who hosts this page also wrote an article in ~October 2002 in CircuitCellar magazine discussing how to connect the Mega128 (in an Atmel STK500 kit) with this proto board.

You are spot-on in your reasoning for switching controllers, although technically the failing is in the RTL8019AS' ability to auto-negotiate. It's just not part of the 10Mb spec.

While the 8019 can be forced into full-duplex mode to eliminate collisions, detecting whether the switch is in FDX mode is a problem. And this is also one of only two 8019 settings you can't elect through a register - you have to install or emulate the EEPROM.

Cheers, Richard

I got the Xport last week from Powercorp in Sydney. Very cool stuff. Soo small. They are selling them as hot cakes and I was told I got the last one for now.

network

series

auto

of

the

one

something

The best one i have seen latey is the lantronix part that has the whole lot in one RJ-45 socket. Good if your only doing a few, probably not so good if you are an OEM. Soanar (polykom) do these parts. Not sure of the cost, but I'm guessing they are probably not to badly priced.

Other solution is to use a NE2000 NIC. Dead simple!

Noise margin.

A device designed for only 10 Mbit/s can have a much narrower frequency response without destroying the eye pattern (ISI) than 100 Mbit/s. The narrow frequency rejects much of the high frequency noise that might corrupt the 100 Mbit/s link. A narrow transceiver bandwidth also reduces cable radiation, which can be important in some situations. Don't use any data rates higher than necessary.

Paul

See the following URL for details of a design using ASIX Ax88796.

regards, Johnny.

The list price for the Lantronix Xport is $49 US. Been thinking about dropping one of these into my 1xrtt RTU project.