1

Hi I wish to learn more about NICs. And I would really appreciate any answers or resources you could point me to! We have a small cluster at work, and I'm thinking about getting 2 to learn with at home.

1) On our 10G network at work, would a 40G NIC make any difference? I read online 40G means 4x10G, and this makes me think it's not "true" 40G?

2) What are the dual-port NICs used for? Could using only 1 of the 2 ports lead to better performance than using both? Does a dual port 40G card mean 80G, provided your switch is set to handle that.

3) I have 2 Linux desktops at home, and I want to install my own driver, and try some tests. I will put in 1 NIC per desktop. I don't have a switch or router. Would a single QSFP+ to QSFP+ cable be able to directly connect these 2 computers for UDP/TCP traffic? I read about crossover cables, but these appear to be copper-only.

4) Is copper "slower" (assuming < 5 feet distance) than optical? I want to mention we use InfiniBand/RDMA at work and so I'm trying to measure performance on scale of nanos, so if copper is slower by >100ns then this is not feasible for my tests.

Thinking about the Chelsio T5 series, and using their provided userspace driver.

Thank you so much!

Edit: if this helps with how you want to structure your answer--I have had elementary exposure to networking and hardware at work, but I am a full time programmer. I have not experimented with device drivers, hence my interest in this area.

closed as off-topic by Rex, MichelZ, xeon, mdpc, user9517 Apr 24 '14 at 6:48

This question appears to be off-topic. The users who voted to close gave this specific reason:

  • "Questions must be relevant to professional system administration. Server Fault is dedicated to professional system and network administrators. End user and enthusiast questions are off-topic (contact your system administrator or hire a professional to help you out). Please see the Help Center for more information." – Rex, MichelZ, xeon, mdpc
If this question can be reworded to fit the rules in the help center, please edit the question.

  • 2
    Whoa, since when did 40G, Infiniband and SFP+ mean "beginner"? Mike, I'm not totally sure we can answer all your questions here as some of them are rather broad and debatable. You might want to break your questions into separate posts. – Stefan Lasiewski Apr 24 '14 at 5:34
2

On our 10G network at work, would a 40G NIC make any difference? I read online 40G means 4x10G, and this makes me think it's not "true" 40G?*

When you speak about 40G, you usually mean "real" 40G (one port) see Wikipedia
You can do port aggregation with 4x10G links, this is also possible

40G on a 10G network is not possible, the NIC will negotiate to 10G at best. Using port aggregation on multiple nodes on the same switch, or between switches with port aggregation, would benefit you eventually though. (Depends on traffic pattern)

What are the dual-port NICs used for? Could using only 1 of the 2 ports lead to better performance than using both? Does a dual port 40G card mean 80G, provided your switch is set to handle that.

Dual port NIC's just have two ports on the same card. Nothing fancy. Using just one port can lead to better performance for the one port if you'r constained by your BUS type (PCI-X, PCI Express and stuff - see Wikipedia)

I have 2 Linux desktops at home, and I want to install my own driver, and try some tests. I will put in 1 NIC per desktop. I don't have a switch or router. Would a single QSFP+ to QSFP+ cable be able to directly connect these 2 computers for UDP/TCP traffic? I read about crossover cables, but these appear to be copper-only.

Yes, you should be able to directly connect them, without a switch

Is copper "slower" (assuming < 5 feet distance) than optical? I want to mention we use InfiniBand/RDMA at work and so I'm trying to measure performance on scale of nanos, so if copper is slower by >100ns then this is not feasible for my tests.

If you're absolutely performance paranoid, go with something other than Copper. However, there's so much of the other stuff wrong in your test then (Desktop PC, Home-Setup, and so on...) that it won't make any difference in your home-use-case here

Not the answer you're looking for? Browse other questions tagged or ask your own question.