CCR2004-16S-2S+ SFP+ Interface RX-Drops

RouterOS General
1

Braintrust,

I just this week upgraded my internet via Frontier (US) to 2GB/2GB. They overprovision, so it’s not uncommon to see 2200MB/sec transfers. I’m a 20 year MT user, so this upgrade with my new router should have been a snap.

They deliver on an ONT that has a 10GB RJ45 port on it. I added a Mikrotik “r2” SFP+ to RJ45 that shows temp and link advisements and negotiated speed to connect to the ONT from the CCR2004-16S-2S+ router on SFPPLUS2, and a Mikrotik S+DA0001 cable on SFPPLUS1 to link to the PC in test or the CRS328-4C-20S-4S+ CRS.

Default config from a factory reset (to rule out my normal config with a lot in it). No extra firewall rules, no queues, “factory”. Grab DHCP from the ISP from sfpplus2, hand out 88.x IP’s on the LAN sfpplus1 port. Fast track is enabled. Both ports are linked up at 10GB full duplex (no speed differences between the interfaces as i’ve seen discussed with the 5009 and the 2.5GB port).

Go do a speed test. I’m not quite getting full speed. About 10% less than if I go direct PC to the ONT with the RJ45 SFP and a Mellanox Pro 3 card. During speed tests, downloads hit about 13% cpu load overall, and uploads about 9% overall, looking at profile, nothing is spiking in usage, plenty of headroom.

Looking at the interface stats I think I have found the culprit. I see a that under RX Stats, RX Drop shoots up in lumps of 500-1500 at a pop, with a run of Ookla’s speed test generating 30,000 or so drops on the interfaces each way per test.

I tried making a queue, using both cake and fc_codel at ranges starting at 1500 thru 1800MB/sec - just under my max speed. That resulted in another 20% speed loss, however the drops slowed down some (but didn’t stop), so that didn’t resolve it.

I turned on Flow Control and the drops reduced drastically (maybe 30 per speed test run), and I’d see TX pauses on the interfaces, but once again, I was slowed down to 20% of the speed I can get directly to the ONT.

Reading thru the forums, there seems to be some discussion regarding the CCR2004, buffer size on the hardware queue not being large enough, and other items, but nothing that’s been a solid fix.

This only occurs under heavy load situations. Traffic under 1GB/sec doesn’t generally create the drops, it does but just a few here and there.

Bufferbloat testing from Waveform direct to PC is an A usually, thru the router, B or C.

I realize this is only 200MB of speed loss so not a huge deal for a home lab, but the drops do cause some packet loss when it’s loaded up.

Any thoughts or suggestions? More info I could provide? I was hoping this CCR2004 was going to be mature enough at this point to be solid. I’d post config, but again, it’s the factory reset default config.

Thanks much,
Dave

2

[quote=challado post_id=1020042 time=1692445050 user_id=27152]
Unfortunatelly,
[/quote]

SPAM

3

No. I have posted with the hope that something in Mikrotik read these. As you can say, several users post, without answer…

4

Please do not post mutiple times the same topic, that is considered spam

5

No. I have posted same post, in different topics, that are excluded. Perhaps anyone are annoying with my sincere opinion, because simply have EXCLUDED my post and another topics in forum.
But OK, it won’t change the reality: product have bad project and bad design, only with low price and nothing more.

Follow the removed post.
Unfortunatelly, Mikrotik’s hardware starts to lost quality in the last years: products with poor design, processing power but the switch architecture doesn’t follow the same evolution, etc.
I have a residential net with two fiber gigabit internet connections (with two different operators, for redundancy and load balancing). Both use DHCP to set valid ip address on a linux router with two intel I225-V and one INTEL E810-XXVDA2 on pci express x16, connected to CCR2004-16G-2S+ 10 GB port. When traffic is constant on 2 GB, a LOTS of RX DROPS are received mikrotik’s 10GB port sfp-sfpplus-1. The marvell chips are problem. When mikrotik use athero’s chips on earlier equipmen’ts I never see these problems.
Now I bouth a TL-SG3428X to simply substitute these equipment with poor design and very very bad projected. I will never suggest anymore to anyone to buy these device.
Also, I have a hAP ac³ thas, as sayed “have a high gain antennas and long coverage”. Another bad designed product, with low coverage, etc. Another product that I never suggest again and now subsitutude by tplink GX 90 device.

6

this is a user forum, if you have some issue with your MikroTik device is a good idea to request to support

https://mikrotik.com/support

off course at the forum you can share your experiences with other users, sometimes that can help

also keep in mind is saturday “a non business day” so you don’t expect an expedite response