3 ms·
> ironically people have been complaining more about Intel NICs than Realtek recently but that’s anecdotal still IMO I'm not sure if Realtek is getting better,
by toast0 4y ago
> ironically people have been complaining more about Intel NICs than Realtek recently but that’s anecdotal still IMO
I'm not sure if Realtek is getting better, I've heard a lot of issues with Intel 2.5G nics; people say some revs of the hardware are unusable at 2.5G (but do work fine at 1G). Personally, I've only got good things to say about Intel 1 and 10G nics I've worked with.
For me, Realtek 1G works ok until it doesn't, it's possible to overwelm the hardware with traffic and then you get bad behavior --- queues get stuck or seem to get stuck and the queue processing doesn't always fully reset when you tell it to --- this can lead to wild writes, following descriptors that the OS thought were dead. Sometimes better results with Realtek's drivers than open source drivers, but they frob the hardware with crypticly named defines, and I've seen them send pause frames (which nobody wants) with no way to disable. On one machine, using Realtek's drivers results in immediate kernel panics, so it's not universally great. The interrupt design is poor anyway --- MSI-X dates from 2002, they could have used a separate irq for rx, tx, and misc, rather than a single interrupt and a status register. But that might be ok if the status register worked properly. It's too easy to have timing issues when poking the status register and then you don't get interrupts you should. I haven't touched their 2.5g hardware, hopefully they fixed some stuff.