Hi,
I've spent a few years playing with routerboards and used a few in real network applications. I now work at for relatively large ISP and thought of using CCRs to offload a few tasks from high-end routers not performing as well as advertised.
We've been doing some lab exepriments to validate the use of RB1x00 and CCRs as netflow probes and MPLS aware LNS routers, but a few aspects remain problematic.
The hardware design, clean and lean in principle, doesn't quite match our usual requirements. Interfaces choice ain't optimal either.
Of course, the lack of use-case oriented documentation, quality assurance on the software and direct support contracts is also an issue that I hope to work around pretty soon. But let's stick to the hardware side.
* Interfaces
In large carrier hotels and datacenters, we don't use copper anymore. Every single link is built using single-mode fibre (rarely multimode). Everything but the local management interface has to use SFP / XFP / SFP+ slots.
For 10G specificaly, when using colored DWDM modules, XFP is way preferable to SFP+ as it allows for higher power modules required for DWDM metro links.
Also, knowing the Tilera Gx8036 has 4x10G capability, it'll be nice to get something a bit more modular.
My point is to consider (cold or hot) plugable interface modules with the following specs :
* Single port SFP+
* Single port XFP
* Quad port SFP
* Quad port RJ-45
* Switched 10-12 ports SFP
* Switched 10-12 ports RJ-45
* Switched 10-12 ports RJ-45 with passive PoE control
4 slots would be made available on the front side, with a middle-plane that could be a unique feature for every models in the CCR line.
* Power supply
With this design, the rear side would have an replaceable CPU board to allow cold upgrades from 9 to 16 or 36 cores as required. It would also host two power supply slots with the following models :
* 80-240V AC
* 12-60V DC
* 80-240V AC with integrated battery charger and UPS feature (for remote FTTx and radio shelters)
* CPU board
The same chassis could also eventually fit a switching-oriented CPU board with less features than a Tilera based board, to re-use interface modules, chassis and power-supplies for remote points.
The mid-plane would allow activation of only 2 slots for a Gx8009 based CPU board and up to 4 on a Gx8036 based board. I'tll handle console port and OOB management network, chassis monitoring (via an OOB SoC eventually) and local storage (at least embeded flash as a backup boot/conf for the CPU board).
* Chassis
The total depth of the chassis is critical for remote endpoints : many switches (Huawei's quidways, Cisco ME 3x00) are too deep for street cabinets. 40-45cm is the upper limit.
The front side display as proven useless in most use-case. Although local monitoring is nice in critical situations, we alwas have local serial consoles to work on-site and can't afford a local access to the equipment with a weak pin code as the sole access control when in colocation telco rooms. On the other hand, a local (PCIe-based ?) management network (out-of-band) port would be a great adition, so would be a RJ-45/48 serial header replacing the DB9 port.
Chassis construction would have to be a bit tougher : 12/10+ galvanized iron is preferable, if not for mechanical requirements, it really helps to gain trust among the old network engineers.
* Pricing
I guess such chassis could have a way higher price-point than the current CCR line. Considering the features and benefits of such platform, a 2k€ price range for a 4 slot chassis fully equipped with redundant power supplies, XFP and 10SFP boards would still be a bargain.
I guess the economy of scale achievable via re-using modules accross products would help keep good margins even considering the raise of construction costs.
I may even believe my employer would invest some cash in developping such models, at least on pre-orders
What do you think, as users and potential buyers ? Could anyone from mikrotik comment on the feasability ?