A GbE device not able to forward full-rate odd-byte frames

Hi, I have a GbE device that cannot forward the full 1Gbps rate for frames of odd-byte sizes, but can forward the full

1Gbps rate for frames of even-byte sizes. I hope to get other experts comments on this situation.

I'm using IXIA traffic generator to test the GbE device. With the even-byte-sized frames (say, 64, 512), the device can receive full-rate traffic and forward back to IXIA. However, with the odd-byte-sized frames (say, 65, 511), the device can receive full-rate traffic, but fails to forward all the traffic back to IXIA. Overflow happens inside the device.

The nominal traffic generation speed for 511-byte frames can be calculated like the following; 1Gbps / (511 + 8 + 12) * 8 = 235,404.89642

And the IXIA's input rate is 235,405 frames/sec. The forwarding rate of the device is limited to 234,932 fps due to the overflow.

The device company claims that IXIA is generating too much traffic, that is, IXIA has a faster ineternal clock than the device clock. Is this valid claim or not?

Thanks for any comments.

Regards, Jay Kim

Reply to
jaymkim
Loading thread data ...

it shouldnt matter - if the clock is wrong the the generated data would be arriving @ faster than 1 Gbps - 1/5 frame / sec or 1 part per million should be well with allowed clock tolerance

i suggest is you try this with different length tests - internal buffering probably takes up the slack up to some limit.

you have to ask whether it really matters that the device cannot handle that last 0.2% of continuous wirespeed thruput.

i suggest you run the traffic generator with 510 / 511 / 512 byte frames and see if 510 is OK as well - that should point out the anomaly on odd length frames.

FWIW odd transfers inside the device memory may have some extra overhead - but if the design is that close to the edge you should be having other issues.

Maybe they misalign internal frame transfers, so for odd length frames they do 2 DMA cycles per x bytes rather than 1? in which case this is likely a bug or hardware programming issue.

Reply to
stephen

Hmm, from the numbers I would rather guess the device under test can only send even length frames and pads if asked to forward odd length ones.

Reply to
Manfred Kwiatkowski

Thanks for your comments. Yes, it seems that the device has some kind of internal bug when odd-sized frames are injected with the full 1Gbps rate. I tried 510, it was ok, tried

509, it failed.

The device company claimed that Ethernet standard allows 200ppm clock difference (+-100ppm) among different Ethernet devices.

I'm wondering this kind of odd-byte problem is common in Ethernet devices. I think it's not, is it?

Regards, Jay Kim

Reply to
Jay Kim

200ppm is 1 byte in 5000, you are loosing 1 in 500.

Who knows. As it makes no difference in everyday operation you will only notice by looking at the wire.

Reply to
Manfred Kwiatkowski

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.