Image Source: Juniper Networks, DAY ONE: ROUTING IN FAT TREES (RIFT) Introduction Routing in Fat Trees (RIFT) is a new approach to routing created in response to challenges that were being experienced in dense, regular, topologies in large-scale data centers. To achieve the goals, more information was required, which meant new information elements, such as Topology Information Elements (TIEs) and Link Information Elements (LIEs). This leads to, from a standard English perspective, delicious sentences like:
Your blog here is so good made me think a bit so here's more. 1. The True LIEs acronyms are fully intended, it's actually quite deeply rooted in psychology/brain physiology and how people learn that led me to choose acronyms that way ;-) 2. SR & RIFT: yes, SR is a great technology to start in the leaf as overlay as e.g. SRoUDP and terminate on DCI, we see lots of that. RIFT will not preclude SR in the fabric but frankly, I have to see a use case for SR in IP fabric underlay I believe in (footnote: it's overkill for error triangulation e.g.). 3. yes, I paid lots attention to the gripes of people actually cabling fabrics, lots interesting things like "I just want to know WHO as in switch/port is on the other side" led to lots stuff you find in the protocol design. 4. Automation/Intent is _great_ and super valuable when we talk about "service monitoring & provisioning". IGP is just standard pipes, e'one standardizes on laying them same way by now (that's your other argument, we see lots of CLOS now in access/sattelite access/core form factors, you name it". The reason is that CLOS is imply the most effective way to interconnect crossbars, we know that since 50s and the math is very hard to argue with. So, use controller to provision service which is unique to every network (since that's the produce you sell while fabric is just "ram chips") but using controller to provision your IGP while you have no IGP is kind of interesting but really unnecessary exercise AFAIS. 5. RIFT is AFAIS the simplest solution to pull servers into underlay with minimal resource consumption and once you multi-home them L3 all the service migration and other headaches go away. True multi-home L3 is not just "download the default route", it needs proper dealing with failures upstream or "on the other side" as well so simple static 0/0 is good neough if you think "blackholing is ok, we have pagers for that" only. And yes, you can always find a more complex way to solve the same problem ;-) 6. I don't follow your argument really how RIFT design impacts BGP, it's just another way to get your IGP next-hop really but that would be an interesting discussion.
Your blog here is so good made me think a bit so here's more. 1. The True LIEs acronyms are fully intended, it's actually quite deeply rooted in psychology/brain physiology and how people learn that led me to choose acronyms that way ;-) 2. SR & RIFT: yes, SR is a great technology to start in the leaf as overlay as e.g. SRoUDP and terminate on DCI, we see lots of that. RIFT will not preclude SR in the fabric but frankly, I have to see a use case for SR in IP fabric underlay I believe in (footnote: it's overkill for error triangulation e.g.). 3. yes, I paid lots attention to the gripes of people actually cabling fabrics, lots interesting things like "I just want to know WHO as in switch/port is on the other side" led to lots stuff you find in the protocol design. 4. Automation/Intent is _great_ and super valuable when we talk about "service monitoring & provisioning". IGP is just standard pipes, e'one standardizes on laying them same way by now (that's your other argument, we see lots of CLOS now in access/sattelite access/core form factors, you name it". The reason is that CLOS is imply the most effective way to interconnect crossbars, we know that since 50s and the math is very hard to argue with. So, use controller to provision service which is unique to every network (since that's the produce you sell while fabric is just "ram chips") but using controller to provision your IGP while you have no IGP is kind of interesting but really unnecessary exercise AFAIS. 5. RIFT is AFAIS the simplest solution to pull servers into underlay with minimal resource consumption and once you multi-home them L3 all the service migration and other headaches go away. True multi-home L3 is not just "download the default route", it needs proper dealing with failures upstream or "on the other side" as well so simple static 0/0 is good neough if you think "blackholing is ok, we have pagers for that" only. And yes, you can always find a more complex way to solve the same problem ;-) 6. I don't follow your argument really how RIFT design impacts BGP, it's just another way to get your IGP next-hop really but that would be an interesting discussion.