So recently my cable modem ( Arris SB8200 ) gained LAG..
I created a bonding interface and set it for 802.3AD with 2 ports.. Im using a CCR-1009..
Everythign works. I see the data spread out across both boths..
BUT… My speeds went way down.. I have gigabit service and I speedtest.net at 950Mbps down.. With both links ( wires ) hooked up I only get 250Mbps… If I unhook one wire I get 950Mbps again.. So with one link I get 950Mbps with both links I get 250…
This seems like a really odd result. A LAG 802-3AD interface on a CCR-1009 can do 950Mpbs to one computer right ? Its not limited in some way like no fast track if I use a bonded interface right ?
I should be able to get 950Mbps on a speed test thru a LAG 802.3AD right ?
I’m afraid that bonding is hardware offload only in crs3xx series.
In CCR series bonding is done on software. But I’ve reached 2gbps with bonding on ccr1009 in a transit router. I mean, different IP addresses for source and destination.
So there should be a problem in your setup.
So my config is that im using Ether1 and Ether2, in a bonded 803.2AD group. I have played with all the hashes and for the speedtest.net that I am using for testing ( or iperf public servers ) the layer 3/4 seems to have the best results for hashing. I test out the same speed using any test method, 275-280Mbps and my ether1 and ether2 interfaces show 140/140 Mbps..
This might be a cable modem issue. I just need confirmation I am not doing anything wrong and I should be able to get 980Mbps thru a 2 port 802.3AD group using the above config ?
This is from a Speedtest.net test.. You can see its spreading out the load for download and upload nicely across both ports. But they only reach 150/150.
My CPU only shows a 1-2% increase during these tests.
If I unplug either cable the speed test will reach 980Mbps.
So to wrap this up, the only way LAG/LACP 802.3AD works at high speeds speeds is to use ports that do not go thru switch chips to CPU via 1Gb internal link ? Thats correct right ? So Mikrotik might want to make that more clear in the specs ?
There is also the matter of how fast you can go thru the combined ports. With 10 LAG members thru a CPU can you get 10Gbps ? What pps ? Does this vary with the number of cores ?
SO… I think Mikrotik should add a line to specs to show what can be done with LAG as its not easy to figure out ahead of time without testing it.
Acording to https://wiki.mikrotik.com/wiki/Manual:Interface/Bridge#Bridge_Hardware_Offloading only CRS3xxx series supports bonding with hardware offload. It means that other devices do bonding with CPU.
CCR1009-7G-1C-1S+PC has 9 cores so it can take 1 Gbps from each G port and bond them in total of 7Gbps in SFP+ port
RB4011iGS+RM can bond two ports from ether1..ether5 and another two from ether6..ether10 and bond them in total of 5 Gbps that can be transmitted/received over the SFP+ port.
RB3011UiAS can agreggate up to 2 Gbps but it will depend a lot on CPU’s usage and port selection. I wouldn’t expect too much. This router is not intended to bonding.
RB2011 can bond 5 Ether ports of 100 Mbit and join the traffic in 1 Gbps port or split 1 Gbps in 5 ports of 100 Mbps.
If you want bonding, as it is a layer 2 work, better choose a good switch that can do it by hardware. CRS3xx works very good. And you can configure it with SwOS or RouterOS.
I don’t agree with you in MikroTik should explain what to expect with LAG… but it will be very CPU-dependent.
Wow, that’s interesting, but I’ve done those bondings using the balanced-xor mode instead of 802.3ad using the first 4 ports of the CCR1009, that config works great it goes up to 3.5Gbps (real traffic) being forwarded between two Mikrotiks (CCR1009 + CRS326) → that’s because the lack of 10gig ports at the ccr, and the excessive amount of ports at the CRS317, my solution is to bond these two mikrotiks and I have three 10gig interfaces with a 4gig bonding in the middle.
Hmm… Yes I guess i agree.. Mikrotik should not work on publishing bonding results as that would get really complex.
The reason im bringing it up is that all new Cable modems for DOCSIS 3.1 are going to come with 2 ports and 802.3ad. Cable companies are going to 2Gbps and to 10Gbps “multigig”.. A ton of non technical users are soon to realize normal consumer routers don’t have the horsepower or LAG WAN capability. So Mikrotik is sure to see more sales to these power users that do not have the technical skills to figure out which models can do it.
So I was just thinking out loud that it might be good to simplify the ability to choose DOCSIS 3.1 Multigig capable LAG routers.
Hello, I hope I’m not drudging up or hijacking an old thread, but it seems as though a few months old isn’t too far back. However, I’m curious, how can 2.5Gbps be achieved on a 1Gbps chip? I currently have a 4011 with Ether1 and 2 bonded and am seeing a clear bottleneck. My graphs show a plateau at 1Gbps. This seems like it could be the issue, but not if 2.5 Gbps is possible?
EDIT:
Looking at the specs for the block diagram, the 8367 chip supports RGMII, which is 1 Gbps, whereas the 8366 and 8369 support RSGMII which is 2.5Gbps. Could this be an issue?
Also, IF my problem is in fact the chip can only do 1Gbps, can I lag ports from each chip? For example, can I lag ether1 with ether6? Thank you for your help!
Depending on bond mode it may not be possible for single client connecting to single server to use both physical links. Meaning that many times, single link speed will be maximum achievable speed. Bond will see full use when it’s used for connections from different clients to different servers, in that case traffic will spread (almost) evenly.
