Hi there,
i spend lot of money for CCR1036, but it seems too buggy.
In my scenario is router with 6.6 ROS conected with ether1 to ISP with 360Mbps up and downstream limit.
On ether3 i have about 15 vlans.
On each vlan are about 1-3 IP subnets with /30, /29, /28, /24.
Other interfaces are unconnected.
In router is about 150 routing records to some IP in vlans.
No firewall, no accounting, no queueing, no conn. track., just absolutely clean router.
If i put router to production environment (large wireless network with many Access points), it looks pretty good. CPU usage about 0,1%.

But absolutely randomly router slow down traffic, and for example ping with packet size 1400B won't work, smaller pings are OK.
After few seconds everything works well again, but it all repeats randomly.
In log are no messages (in debug level too).

In attached photos you can see one slow down, and on second picture one slow down, and second much longer.
If i connect my old linux router with the same configuration, everything works well.
It looks like some bug in 6.6 ROS.

Can anybody help?

Sincerely

Martin

One more information,
in the ARP table is about 93-95 record.

What testing method did you use?
Traffic Generator?
Torrent download?

Hi, we are also having problems with the CCR1036.

We are an FTTH (fiber to the home) ISP and we have deployed a Cloudcore CCR1036-12G-4S in our data center as a main router from our customer networks out to our upstream internet provider.
The Cloudcore is only doing basic routing between our public IP range and our backhaul provider public IP range. On the uplink we have a 1Gbps breakout that reliably gets about 600Mbps to any site within our country, and around 200Mbps international.

The problem is that the Cloudcore starts to add a delay to establish connections as soon as the load from our customer sites exceed 25Mbps; and roughly 16000+ connections. Especially HTTPS sites become very slow to respond at first (10 seconds). Once the connection is established it works fast.

On each FTTH site we have RB1100AHx2 routers acting as our BRAS systems, PPPoE servers, doing NAT from private range onto public IP. They all link over a Layer-2 MPLS into the Datacenter where they connect into the Cloudcore Router. The Cloudcore is then only doing basic routing between our public IP range and the upstream provider public IP range.

On the uplink we have a public IP 41.x.x.x/29 which links to a Cisco Catalyst 3750 of our backhaul provider.
On the downlink we have our own public IP range 41.a.b.c/25 which links to all our sites (MikroTik RB1100AHx2). Total of about 1000 customers behind 7 RB1100AHx2 units.

The router has the throughput, we can easily get over 500Mbps. But it has a delay when making new connections? Especially to HTTPS sites it delays to 10 seconds. When I drop all the customers the response is under one second to load the same HTTPS sites. Yet there is ample bandwidth, a lot more than what our customers are using. The CCR1036 CPU sits at 0% and RAM is almost empty, about 3.6GB available.

Anyone experience something similar?

What testing method did you use?
Traffic Generator?
Torrent download?

Testing method!? Live production environment!

The problem as i described disapear after upgrade to 6.7.

We are using our CCR1036 in a live environment.
We are already on ROS 6.7
The problem is primarily with HTTPS SSL traffic. If the load goes anything over 15 Mbps the delay in opening any HTTPS site becomes noticeable. When we hit 60Mbps the delay is so bad that some local banking websites take up to 12 seconds to open, and the second time you open it, it is quicker.
We are not using DNS or anything fancy on the CCR1036 - a simple router with 2 public IP addresses.

Hi again.

Yesterday i made another CCR test in live environment without success, and with similar behavior as riaanjvr wrote.

Simple testing scenario: dudeserver-internet-CCR-WISP-myPC.

In case of problem i see:
- router stops response for SNMP questions from dude server, anything in WISP is from internet at SNMP unaccessible (new connections).
- traffic at ether1 slowdown, and on all internal interfaces too (ether3 with cca 15 vlans).
- traffice between vlans is slowdown too
In case of problem i never lost connection from my PC to dude server, but i was unable to start new connection, for example open any web page.
All other WISP clients have similar problems too.

After few seconds sometime minutes traffic goes up, and everything works well.

It looks like a problem with opening connections through CCR (who knows), any established connections works well.
Conntrack table have approx. 20.000 connections, possible is about 450.000. Router CPU is abou 0-2% usage, RAM about 3,5GB free

For problems above is opened ticket #2013121166000078.

--
Martin

Was there any follow up to this? We are passing 300mbit on a CCR with 6.10 and getting many reports about SSL timing out, etc. No issues with our x86 routers on a different network passing the same amount.

I'm having the same problems and i'm on 6.18.

Do you have any solutions?

I have no time to play with this and a don't want to make angry hundrets customers, so from my last post till this time and may be forever CCR1036 lasting as a brick.
Very expensive brick.. Sad but true.

from the looks of it you are still using the old and buggy routerOS. Try the new one and see if it still persists. I'll need more info to replicate this as well.

I seem to be experiencing the same issue with my CCR3036. I can not attribute this to a load issue though.

It seems that at any time of the day, SSL sites have a large delay initiating TCP transmission. Many banking sites are extremely slow and some practically unusable.

Any insights into this from the Forum?

CCR3036 ROS6.17

Ok. False alarm.

My particular issue seems to be tied to an errant Jump in my firewall rules.

Im sorry for digging this old thread but is the problem solved or the culprit identified? Im experiencing same dilemna with my RB750gr3. TIA