CCR1009 CPU bottlenecking without jumbo frames

Hello. I just noticed that CCR1009 seems to bottleneck on CPU when sending data between 2 servers where one has MTU 9000 and another has 1500. It bottlenecks on single core (all cores are 0 except 1 that is 100%) and transfer is capped at around 23 MB/s. After setting MTU to 9000 on both sides load is fine (up to ~15% peak worst core). So are transfers. It makes me worried about 10G performance even with jumbo enabled if it struggles with 1G without them (assuming it’s gonna scale more or less proportionally 9000/1500 = 6 which is the same magnitude as 10) :C I’m talking of course about single TCP connection for huge data transfer tasks, not aggregated throughput. Average traffic is low here, what I really got this device for are those occasional data synchronization transfers between machines in various firewall zones.

I have:

  • plenty of firewall rules (around 120 total but this path should hit ~ 20 rules maybe less)
  • disabled fastpath (because I use bridge firewall inside single network for certain devices)

I actually hoped for hitting around 10G with SFP+ for data transfers with jumbo but now I’m not really certain if it’s gonna happen :confused: Is it normal for this router?

False alarm - I forgot CPU has been downclocked to 400Mhz because performance used to be sufficient anyways…

Yes, it is normal in this case. The CCRs use the Tilera CPU: a bunch of “weak” cores, that spread the load. They aren’t a good choice if You want to get a high throughput with a single connection.

This scenario would be better served with a CHR (where you can throw a mighty x86 at the problem), or with another brand router - one that uses a CPU with less cores, but more powerful ones.