We are working on some future high speed routers with around 100Gbit throughput, and we would like your input on what types of ports you would want to see in such devices.
If you imagine a device with around 16 ports (just an example), how many of them would you want to be (just examples, you can add your own ideas):
SFP+ (10G)
SFP28 (25G)
QSFP (4x10G)
10Gbit copper (10G)
Management with PoE
Let us know in the replies here
Only some ideas:
Router A: 16x SFP+
Router B: 16x SFP+/16x RJ45 10GbE (combo ports)
Router C: 8x SFP+/8x RJ45 10GbE (combo ports) + 2x QSFP
:
:
later
(Router X: 24x SFP+)
(Router Y: 24x SFP+ plus 4x QSFP)
(Router Z: 24x SFP+ plus 4x QSFP28)
(Speeking for myself: I do not need 10GbE copper at all as 10GbE SFP+ optics are cheap nowadays)
all with:
I didn’t know what ASIC you are looking at, but I’d prefer an ASIC/SOC with less CPUs (1big one instead of 72 small ones) According to that slide from Cavium, the biggest highend ASIC is only 500$, so it would be absolutely possible to build an inexpensive highend router, although this means multiple CPUs and therefore single thread performance isn’t that great.
The Indigp NPS-400 Network Processor by EZchip looks better. It looks like it has less CPUs. I cannot find pricing information, unfortunately
But you talked about 100Gbit/s throughput, so the NP-5 seems to be the right one:
I think that would be a great idea some group of port be on a expansion slot so we can change the type depending on our needs.
Enviado de meu XT1580 usando Tapatalk
What format do you mean?
I think that would raise the costs for those routers too much. Fixed configurations would be a less expensive solution. 1x product design / mainboard with different ports attached to the ASIC.
Router with 24 port? Really it not a over port config for a router?
I think a version with expansion slot is not bad, it will be a super core router, I don’t think that will make super high cost on the product
Enviado de meu XT1580 usando Tapatalk
You can already use a high end PC as your router.
But do you agree that is not they same of a dedicated hardware for this
I really don’t think that someone will build a high end server to be placed on the the core high density network traffic
Enviado de meu XT1580 usando Tapatalk
The advantage of specifically manufactured hardware is that manufacture can test it at the specific configuration. If you take that away, is becomes the same as any other PC where you have no driver support.
Primo.
Is it possible to just make funcionality of stacking Mikrotik’s devices ? High-end ones. Then we can buy models with proper ports and stack them. No need for zyllions versions …
Secundo.
One mainboard and slots for different daughter cards … are we going to be CISCO/Juniper like ? 
And this is a problem?
Enviado de meu XT1580 usando Tapatalk
What’s wrong with giving the big boys a run for their money? 
In an other post, I had suggested a chassis with slots, like the Cat6500 family.
In previous MUMs, consultants have talked about replacing Cisco with MikroTik boxes.
I think MikroTik is going on the right direction.
Stackable is also a great idea. Grow your network with your business.
Off topic, MSTP (pvstp, 802.1Q) pleeease! If you’re going big, go all the way.
Sent from Tapatalk
Not at all… havn’t you spotted ?
Stacking could be even the first step to HA funcionality which is missing for ROS.
I dont think high end routers should be based on Cavium processors.
They have low freq upto 1.2Ghz and some specific RouterOS tasks require higher clock speeds such as simple queues, BGP, PPP etc..
So unless RouterOS is highly multi-threaded in such scenarios like to be used as a PPP concentrator etc.., I would prefer some different architecture than Cavium which will provide higher clock speeds instead of super-high multiple cores with low clock speeds.
Ah ok
I thought that you was complaining…
Ha and some kind of Hsrp for mk
Enviado de meu XT1580 usando Tapatalk
New inexpensive member to the CCR family of routers:
No port sharing / TILE-Gx based CPU
12 x SFP+ (10G)
4 x QSFP (4x10G)
1 x Management (no PoE)
1 x serial management
48 port Gb/e and 4x 10Gb/e (SFP+) (TileGX based)
24 port Gb/e and 2x 10Gb/e (SFP+) (TileGX based)
16/32 port 10Gb/e (SFP+) (TileGX based)
The above NEED to be able to do line rate for IP/IPv6/MPLS traffic, or close enough to it (like the way UBNT does by using an Octeon ASIC to get a decent level of mpps, although UBNT doesn’t do MPLS at the mpps level. We need MPLS at that level too).
Or please support the following chipsets from Broadcom. The Trident II and the Tomahawk.
https://people.ucsc.edu/~warner/Bufs/StrataXGS_Trident_II_presentation.pdf ← Trident II
https://www.broadcom.com/products/ethernet-connectivity/switch-fabric/bcm56850 ← Trident II
https://www.broadcom.com/products/ethernet-connectivity/switch-fabric/bcm56960 ← Tomahawk
Lastly, if one were to compare to say Juniper or Cisco I would say the comparisons would be…
A 'tik equivalent to the Juniper ACX1100, or an equivalent to the Cisco ASR-920-12SZ-IM.
A 'tik equivalent to the Juniper ACX5048 would also be fantastic.
Personally, I’d be interested 2x QSFP and the rest with SFP+ and the obligatory Management Port paired with a rs232 for console access.
The pair of QSFP would provide redundant access to e.g. core switches, while the SFP+ would connect to ISP, DMZ, or further routers/firewalls/gateways.
In regards to the PoE-in functionality, I very much doubt that even PoE+ will provide sufficient power. PoE++ / 4PPoE on the other hand just might. I’d much rather see a pair of redundant hotswappable PSUs with a bit of headspace.
Speaking of hotswappable: I think we’d all love hotswappable fans.
Grüße,
Philip
2x QFP28 (with breakout support), 10x SFP+
2x QFP28 (with breakout support), 3x QFP (with breakout support)
3x QFP (with breakout support)
QFP28 Can do 1x100G or can be broken out into multiple 25G.
QFP can do 1x 40G or can be broken out into multiple 10G.