Current model CCR's (even with 10GBit interfaces) appear to be limited to single CPU core per single TCP stream. A CPU core seems to be able to handle around 1gbps traffic, ergo, any single stream appears limited to 1gbps.Why wouldn't it be able to do this?Only if it can do > 1Gbps tcp stream
Unfortunately it is conjecture - but it is a major performance flaw related to the multi-threaded nature of the Tilera CPU's. ROSv7 has been stated to be written in mind of utilising the multi-core architecture, a lot of this talk has been more around processes such as BGP, but one would think offering 10gbit interfaces on the 36 core, and the models we've seen of 1072 in the wild having 8 10gbit ports (if I remember correctly) they would certainly need to support multi gbit TCP streams.Joe,
Have you had it confirmed ROSv7 will fix the 1Gbit per connection issue ?
Just because its not current usage requirement doesnt mean it should be limited to that. It is obvious what the problem is, and fixing it now before its a real problem for the general use is a good option. 1 valud example that comes to mind is perhaps private x-connect to another provider in the same DC, for backup or DR purposes.Sorry, but i do not get these talks about 1Gbps+ TCP connection
Where do you have actual use for this?
For regular size packets (MTU/MRU 1500)
Latency have to be in microseconds to get such TCP connection speed, so it is impossible over Internet anyway
OK, one can use it in server rooms, but you can easily do it with 10G switch there.
Only possible way to get such TCP speeds is with large size packets (MTU/MRU 65535), but currently it is also impossible over Internet, so we are back to server rooms
Maybe in future, but at this point in time this is hardly the biggest issue.
Just because its not current usage requirement doesn't mean it should be limited to that. It is obvious what the problem is, and fixing it now before its a real problem for the general use is a good option. 1 valud example that comes to mind is perhaps private x-connect to another provider in the same DC, for backup or DR purposes.
This is one reason why SFP+ can be good, the port can basically be what you want it to be, from SM or MM Fiber to Ethernet and other possible SFP+ modules, very versatile and a good option for such large core routers - if you aren't at the point of Fiber Optic x-connects and links, then the 1072 is probably too big for your needs (could be exception to this). This is why things like the 36 core exist, with its configurations of ethernet and SFP/SFP+ portsCan't wait for this.
I do wish it had more than one Ethernet port, but I guess you can get there with a SFP to copper ethernet adapter anyways.
This is in my eyes not so really interesting, caus of the power of this router!any information for the power consumption?
That is less to do with the 1072 and more to do with ROSv7 (or software in general).Any news on the multi threaded tcp forwarding.
Is it still going to be single cpu and bound to 1Gb
Awesome! When I visited Riga (it was more than 10 years ago) I bought sweater with such pattern for one of my friends. But it was a bit more colourfull.We certainly started from thereThis ventilation cuts on the top look so extraordinary. Am I wrong or you are imitating some pattern used on national cloths?
Why to replace switches by routers?Teaser not too long now
We are looking to replace 10 of our core Cisco switches with Mikrotik CCRs. The CCR-1072 would be perfect but we need to know when they will be available. What is the value of "not too long" in calendar months?
review shows pre-production unit. final version will have slightly different internals.Here is a review of the CCR 1072 with Pics
http://www.stubarea51.net/2015/07/10/mi ... use-cases/
From the manual:Can anyone confirm that the 10Gbps interfaces can run the 1Gbps fibre modules?