I have placed 3 RB1000 units in a co-lo facility where we route IP space to. Two routers are designed to operate in a VRRP pair as a transit router gateway. The third RB1000 will sit behind this pair and operate as a server router and firewall. At present the VRRP system is not configured and we are operating with a single router

Transit --- RB1000 (head01) --- RB1000 (client01) ---- host servers

We have a single L3 switch (Netgear) with 48 ports and these are configured in various VLANs. The switch ports tend to be left in 1000mbit autoneg.

The head01 router connects to transit using a dedicated /29 link network. This is connected on ether1 in a VLAN (id 61). The head01 and client01 are linked using a /29 link network. This is connected on ether1 in a VLAN (id 4090).
The client01 connects to host servers using several routed networks. These are connected on ether1 in a VLAN (id 4093)
As a backup we connect ether4 of each router in a back channel VLAN (id 4093) using a different IP space

The routing is simple. We configure transit to route a network to head01. head01 subnets this and routes to client01. client01 terminates this range for host servers and provides gateway. Packets at host server use client01 as gateway (in their relevant subnet). client01 uses head01 as gateway over link network. head01 uses transit as gateway over link network.

All link networks are configured such that each router publishes into the /29 and use each other as relevant route targets.

This all seems simple and it should be. However the performance of host servers to transit is awful. It is slow and very jittery indeed. It barely sustains 120KB/s - and we have upstream of 2x1Gbps.

Using an alternative router (not RB) we can sustain over 10mbit/s to transit with perfect performance - no jitter etc.

I also saw something that worried me. ether1 on head01 would randomly stop routing packets. A reset of the router fixed this but it happened again - with light load and within 24 hours. I found that if I disabled the ether1 port then enabled it again it corrected the problem and packets flowed again. I then disabled autoneg on the switch and RB1000 and this stopped the issue.

I am not sure what is going on but have several ideas and would welcome any comments:

1. The use of one port for through routing (in effect using RB1000 as an inter VLAN router) places undelievrable demands on the hardware.
2. The configuration of the routers is incorrect.
3. Compatibility between RB1000 and Netgear switches
4. Some odd behaviour on switch ports e.g. autoneg

To test 4 which I thought was most likely cause I set the switch ports to 100mbit fdx and disabled autoneg. I did this at RB also. This did not improve the situation though stopped the port stopping issue.

I have not yet been able to test 3 as we only have Netgear switches in the centre. However we have a range of other routers and servers and they do not show the same problems.

I have deployed simple configurations to test this and have deployed a very basic configuration. This did not appear to correct the problem. All configs were deployed using a basic minimal setup script uploaded and imported then completed using WinBox.

I have not been able to test using two physical ports. However given the extensive support for VLAN in RB I would have suspected that inter-VLAN routing, especially on RB1000 destined for rack environments, would be typical usage and represent a case that had been tested.

Has anyone seen a similar issue or can suggest anything? All help would be appreciated. At present we have a stock of 12 Rb1000 and 50 Rb450 and this has caused a slight concern - to say the least! Thanks in advance and I look forward to any comments or feedback.