Revisiting IPerf Testing on the Mikrotik CRS504-4XQ-IN 100GbE Network Switch

preview_player
Показать описание
Featured Products: (affiliate links)

Chapters:
00:00 Introduction
01:05 Server Setup
01:28 IPerf Tests

Contact Info:

Disclaimers and Statements:
► I receive a small commission on purchases made using my affiliated links shared the video description and comments section. The views and opinions expressed here are my own, unbiased, and not influenced by this commission in any way.
Рекомендации по теме
Комментарии
Автор

Please let me know what you guys think of the switch and testing!
(affiliate links)

HomeSysAdmin
Автор

You're getting a lot better at making these videos. Another job well done, sir. Keep up the good work. I notice you're already at about 2k subs, nicely done! 😎😁

korishan
Автор

Great Video! Found it very much interesting, especially how you explain your setup and what command's you used

edwynstapel
Автор

first review on this item on youtube - congrats - this is not unobtanium for smb - it is probably the most sensible upgrade they could make - good roi if you value fast in house compute and time management - how fast does this go in real world - matches up well with nvme raid...please do more followups as you start using it a bunch

shephusted
Автор

Pretty cool. Nice benchmark. Thanks for sharing

loucinci
Автор

Just discovered your channel after picking up a pair of ConnectX-4s. Good stuff! Considering picking up this MikroTik switch next, but I heard it's missing the necessary L2 and L3 congestion control features for RoCEv1 and v2 respectively. But I also heard that with RoCEv2, it might be OK anyway. I'm just getting into RDMA stuff, any chance you could give it a try through the switch and let us know your results? Thanks!

devplayer
Автор

mikrotik tests with some IXIA or similar boxes and puts the results on their site, including packet per second count at various packet sizes

you can pass -w1m or whatever size TCP window you want to iperf and it will transmit with a larger TCP window which will lower the need for parallel threads to max out the capacity

chriscappuccio
Автор

100Gb is insane for a home lab. I'm excited to be upgrading our core routing and switching in the data center to 40Gb, though many data centers are deploying 400Gb links nowadays.

paul.phillips
Автор

Awesome switch! ^^ Would you be able to do a routing (BGP preferably) test? I'd love to use it as a spine, but haven't seen anyone do it. But be sure to offload the routes to the switch dataplane so that it won't kill the CPU : P

LampJustin
Автор

This is all well and good, but where is the switching capacity on the datasheet? You're only testing a single port but I'm interested in the entire backplane capacity of the switch. It has 4 ports of 400g so the backplane should support 400g line rate being that it's a switch.

itsnebulous
Автор

Enjoying the content, what would be the difference in real use for your application versus the performance testing with Iperf?

Mr-Locke
Автор

Hi good video, do you mind doing iperf tests using this as a layer 3 switch specifically using NAT? The 98DX4310 ASIC in this switch should be able to do linerate routing but I do not think that NAT is programmed into this ASICS pipeline so it is reprocessing packets or punting them entirely I am very curious to see the performance hit for this specific box.

dabb
Автор

I have one of these switches... How with my recent upgrade to fiber internet, I can get the full 500 gbe upload speeds thru it.

wiredhot
Автор

Try with -Z in the client, I got 94gbit/s with that.

ArnaldoMelo
Автор

I have AT&T 1Gb up and down; I have this switch. I have tried every imaginable setting and even test cables, and I can just get 28 MB download and 1 GB up. Do you have any suggestions

nicknick
Автор

why don't use qsfp28 fiber? there are some really cheap 100gb modules out there :) single mode fiber cable is also pennies...

stankobulanov
Автор

Are you saying that a squirrel sometimes finds a nut 🥜

clarencewiles
Автор

Man, I thought iperf3 would be one iperf better than iperf2 — but obviously, iperf2 kicks iperf3’s ass.

jeverett