HI
I am experiencing to many Rx drops (about of 3000 per sec) with a traffic of 100 Mbps. The server ethernet interfaces are Intel PRO/1000 For the moment i have removed some traffic and there no drops .When I have replaced the configuration at one Mikrotik Rb1000 ,no RX drops
The interfaces are connected at one cisco 3750 switch .One (outside ) connect to access-port ,the other interface connects to trunk mode port

Someone has experienced this issue ?

regards
ilir

please, can you clarify when exactly you have problems? when you connect your x86 box with e1000 card in it or with RB1000?

Also, what is link speed and how much data you are pushing through it, what is packet size.

Thr RB1000 has no drops is ok ,but I can not use it due to load .

the server has tw0 ethernet gigabit interfaces and Dual Xeon processor .At the 45 Mbps ,both interfaces have Rx drops (no errors)
Attached is the print screen of the interfaces .
The auto nogotiaton is disabled ,Speed 1000 ,duplex full

Regards
ilir

Same problem here.
I'm using intel onboard NIC, Intel motherboard.
Traffic is about 100 Mbps, 28k pps.
cpu is 15%

Is the buffer of the onboard NIC to small to handle this traffic?

Tonight I'll try RB-44G NIC

Onboard NIC (intel) on Intel mainboard has RX drops problem.
RB-44GV NIC doesn't have problem with RX drops, but has problem with FCS/CRC error on Cisco switch. (tested with several cards)
RB-44G NIC has no problem with FCS/CRC but the RX drops exist.
Dlink 538T has problem with RX drops, but it much much lower than the others.

The RX drops problem only start when you have a lot of traffic, say above 45 Mbps, and a lot of rules (mangles/filters).
My test showed that without rules, just bandwidth test between core 2 PC & RB-450G, has no RX drops and bandwidth reach 500 Mbps.

I left my configuration with Dlink 538T (the best performance), 1700+ lines of mangle, 1300+ lines of queue tree, 20+ VLANs, 100+ Mbps bandwidth.
Dual Xeon 2.26GHz, 8 cores, 2 GB RAM, SATA DOM Level 6, ROS v4.9
CPU below 15%

RX drops still counting, but slower than before (using onboard NIC).
I wonder if I use "best of the best" NIC exists on the earth might solve this problem.

I sleep only 3 hours last two nights to figure out solving this problem, have no best answer yet.
Please anyone that have HUGE traffic on the Mikrotik box (as mine) please share here.
I still believe Mikrotik can handle this size of traffic.

PLEASE HELP

Thanks.

it's a matter of CPU!

Using double xeon (totally 8 cores) 2.26 GHz does not good enough to serve 1200 lines of mangles and 200 queue policies.
Mikrotik use one core only for firewall & queue, that's why the cpu resource shows 15% (one core equal to 12.5% for 8 cores cpu)

I change the CPU to AMD 3.0 GHz and none ... let me repeat ... NONE ... of packet drop on the ethernet, and everything is work as is means to be.

Hi,

What MoBo and NICs do you use?

Regards, Grzegorz.

Hi,

What MoBo and NICs do you use?

Regards, Grzegorz.

Gigabyte GA-MA785GM-US2H
Onboard NIC
AMD Phenom II 3.0 GHz

New find out.

Using Intel Core i5 3.33GHz with equal Gygabyte MB is no better than AMD 3.0 GHz

Using Intel shows counting RX Drops.
Using AMD shows NO drop at all.

i have intel here too and one of the ether shows rx drops 5 in 5 minutes exactly.
the proc is a dual core.

if i set the /system hardware set multi-cpu=no, it can get better?

thanks

i have intel here too and one of the ether shows rx drops 5 in 5 minutes exactly.
the proc is a dual core.

if i set the /system hardware set multi-cpu=no, it can get better?

thanks

no. keep multi-cpu=yes
if your mikrotik detect 2 cpus, just make sure the load is below 50% otherwise it will start dropping packets.
auto negotiation/duplex settings on ethernet interface may cause the drops also.

i have intel here too and one of the ether shows rx drops 5 in 5 minutes exactly.
the proc is a dual core.

if i set the /system hardware set multi-cpu=no, it can get better?

thanks

no. keep multi-cpu=yes
if your mikrotik detect 2 cpus, just make sure the load is below 50% otherwise it will start dropping packets.
auto negotiation/duplex settings on ethernet interface may cause the drops also.

i kept the multi-cpu=yes and set no auto-negociation. The Rx drops stoped, but stil have some problem like the image attached.
this is what happened when had rx-drops.

thanks.

If you use 1gigabit setting to autonegotiation reduces the RX drops substantially. I did this 2 weeks ago and the rx drops and tx drops have reduced to 1-3% of the original.

I stll want to take it to zero. I am seeing rx drops on the brodcom nics on our dell r200s and 2950s but no more on the intel nics. Using 3.27 on 2950 and 5.0rc1 on r200s with intel 1000pro-vt on 3.27 and 82576 quad port based nic on 5.0rc1

I would like to see this at zero as traffic flowing is not more than 150 mbps -200 mbps on the individual nics

After struggling for couple weeks, I found the cause of the drop is the CPU.
I use dual 4 core xeon (makes 8 cores) but Mikrotik only use 1 core (because I only use 1 interface)

Drops happen when CPU (a core) reach 100%, you can see each core usage in ROSv5, in ROSv4 cannot see each core.
In my case, using 8 cores, means each core have 12.5% if you see it on the /system/resource.
If the CPU usage is not much higher than that number, means you only use 1 core.

The solution is to spread traffic into multiple interface cards, each interface MUST have its own interrupt (IRQ) so it will handled by different CPU cores.

No drop since then.

here, i solved it changing all the hardware. With another dual core, i don`t have the problem, and with a quad core too.
I think the first dual core, there was some problem with motherboard.

but what brianchrist said is true, about the cpu. I`m using the Ros 5.0 now, and it`s so good and stable.

thanks all

having the same problem

HP ML 350 2x Quad Core Xeon. Intel pci xpress nics , and only 5% cpu load using router OS 5 rc2 and i got rx drops sad.

Hi,

We have the same problem. Can somebody confirm that solution is to use amd over intel, to eliminate rx drops?

i have no idea how to solve this problem yet, but it's not bothering me anymore. I have changed intel ether for realtek and the problem remains, so i think HP it's not the better one to do this. And now, i can't try another brand.

How about you?

thanks

For now solution is to have more and more pppoe servers as routeros cannot use all cores equally. But there is another problem with more than 2 mikrotik pppoe servers in one network.

Now I have 3 pppoe server in a Switch.
2 HP server and a RB1100. The Rb1100 i can't connect more than 300 sessions because of the cpu load higher than 80% and get 100% when I open some informations.

I have no problem now, just the rx drops in the servers.

having the same problem

HP ML 350 2x Quad Core Xeon. Intel pci xpress nics , and only 5% cpu load using router OS 5 rc2 and i got rx drops sad.

LOL Are you joking or what???

343 "RX Drops" in 9,5h uptime - that is not a problem, that is usual thing on high speed and pps environment.

7 digit number "RX Drops" in 24h - that is "RX Drop" problem, in your case it is just normal driver behavior.

Now imagine situation when these numbers are not available to you - will you actually feel that there are problem. that out of 25000^9,5^60^60 packets 343 didn't even arrived and probably were resented moment after.

Your QoS is probably dropping 1000x more packets and all still is fine.

Macgaiver,

When we disconnect few of our pppoe servers and there are plenty users on one machine with bandwidth 200M+ on one interface there is 4000+ packets dropped every second. When user go to pingtest.net it show packet loss 1-4% and that is not good.

Al machines have 4 physical cores and IRQS are balanced (different irq for different nic) on v4.16
Tried many motherbaords and CPU models (except amd) but it is the same.

I think you're discussing for two different things.

400 TX/RX drops in 10h will never cause 1-4% packet drop. Therefore, you can have no RX/YX drop at all and still have that packet loss, thus they are different things.
BTW, 1-4% of packet loss is no bad (VoIP consider 5% of packet loss and still not detect problems in the conversation.

I think that you didn't understand : )

In our case it is 100 million drops per day not 400 in 10h.

I have also the Rx drops problem . Intel Pro 1000 . These numbers are for 1 day uptime only

The Mikrotik Support only said to analyze this video (I already opened 2 tickets):
http://www.tiktube.com/index.php?video=363

Nothing changed after these informations.

Intel Pro 1000 Rx drops problem too

The Mikrotik Support only said to analyze this video (I already opened 2 tickets):
http://www.tiktube.com/index.php?video=363

Nothing changed after these informations.

Link not work ;(
ERROR!
Requested video not found!

This was a video of some MUM, about cpu configuration in ROS 5.x.

This didn't work for me, but the real problem was bridge configuration. When I changed the root bridge here, this problem was solved (in some places, I changed rstp to none in the bridge protocol).

Sorry for grabbing this old thread out, but i've the same problem with Intel 1000 Pro. Any solution yet?

use latest version - 6.x most probably - it has newer Intel drivers

Update to 6.0rc3 seems to be okay, no more RX Drops for now! Just continue monitoring!

Hi again, for now the Drops are gone. Few packets are dropped, don't know why but 255.000.000/500 Sent/Dropped are okay for me! Maybe the MT-Stuff could implement the new intel driver in an official stable 5.x release?!

Edit:
RX Drops are back. Seems to be related to OSPFv2 bug

Edit2:
Support found the reason for my drops. Related to my NIC Chip (listed in mikrotik wiki http://wiki.mikrotik.com/wiki/Supported ... t_chipsets ) 82541GI

I know this is a really OLD POST, however I have noticed rx drop issues and I am wondering if I can turn on some debugging information in the logging to report why the mikrotik thinks these packets are being dropped.

you guys need to play with queue types... also had that problem with a intel x86 system and intel ethernet
change the pfifo queue size to 2000 ou 4000 pkts (reboot system after changing the size so it can kick in)

i have similar issue RX drops in x86 ros 6.58.5 stable version.

this happen in my x86 dual xeon,
almost all interface with high traffic above 100mbps has a little drop.. sometime increased every day by hundreds ..sometime only tens.. depending on how much traffic running on that time..

however, i have got x86 with single xeon.. no issue at all .. no rx drop.. the ethernet card is same intel i350 and intel x540

i guess CPU issue.. i have not tried further more since rx drop only a little and not causing any serious problem.. but still curious why rx drop appear

Unfortunatelly,

SPAM

i have similar issue RX drops in x86 ros 6.58.5 stable version.

this happen in my x86 dual xeon,
almost all interface with high traffic above 100mbps has a little drop.. sometime increased every day by hundreds ..sometime only tens.. depending on how much traffic running on that time..

however, i have got x86 with single xeon.. no issue at all .. no rx drop.. the ethernet card is same intel i350 and intel x540

i guess CPU issue.. i have not tried further more since rx drop only a little and not causing any serious problem.. but still curious why rx drop appear

Hmmmmm me too... been following this thread from the begining and looks like v7 also has this issues.. on the intel nics.. intel x520da2 and x710da4 also throwing out this errors.. for me is just rx-errors running out of ideas..because RPS is enabled.. i am running 2 intel R620 with 48cores.. 2.7ghz each server.. i can see flow traffic packets beeing distributed even on the cpu.. average 5% to 10% peak times per cpu core.. i also have mellanox cards connect-x4 on the same server.. an no errors at all.. both servers have intel gbic connected from x710da4 to other server intel x520da2 850nm tried several intel gbics but still have errors.. tried playing with multi-etherhet-default with mkpfifo 5000 packets.. but still get errors.. only thing i did not try yet was the bridge information switching RTSP to none.. will try out later stage...

Ok so further testing i have actually noticed something..

the intel X710da4 and the intel X520da2.. somehow on the IRQ ... all queues are being used only on CPU 0 the other 47 cores are not beeing used..

but on the mellanox queues card i can see all different queues beeing spread out across all 47 cores of the cpu.. so i am now 99% sure this is the issue of the RX-ERRORS drops on the intel cards.. this happens both servers we have runnning xeon with 48 cores cpu..

attached picture of i40 card and mellanox card port1 and port2 surely this is the issue happening on the losse.. the CPU is set on AUTO i cannot change it on the IRQ