Re: [Rift] 4 questions（draft-przygienda-rift-dragonfly-00)

[Antoni Przygienda <[e-mail]>]

  1.  I don’t count links but yeah, imagine it’s a multiplane clos in the 2nd figure and count what you miss for it 😉
  2.  & 3) yeah, in figure 3 it’s 2 fabrics with 2 pods with 2 planes (to show the simplest most complex case)

  1.  nah, any amount of links, it has no limit on number of fabrics but you cannot take more than one alternate hop (arguably, that would be a very bandwidth wasting, delay adding idea)


--- tony


Juniper Business Use Only

From: [e-mail] <[e-mail]>
Date: Thursday, 2 November 2023 at 09:53
To: Antoni Przygienda <[e-mail]>, [e-mail] <[e-mail]>
Subject: 4 questions（draft-przygienda-rift-dragonfly-00)
[External Email. Be cautious of content]


Hi, tony

When I read the draft, I got 4 questions. I appreciate your clarification.

1)  In 1. Introduction

"We can see that it is nothing else but a multi-plane CLOS with a lot of broken links for standard RIFT."

Does broken links means: direct links between LB1 and SA1/SA2/SC1/SC2，direct links between LB2 and SA1/SA2/SC1/SC2  and so on？

In figure 1, totally 24 broken links?



2) In figure 2, is Fabric A a PoD, or LA2 and LA1 form a PoD, or PoD is not applicable to Figure2?



3) If we look at Figure 3, there are 2 PoDs each for Fabric A and Fabric B?



4) 4.1. Additional Bi-Sectional Bandwidth Route Computation Chang

Does  this draft limits the sparse DF+ topology  to only 2 links coming out of one ToF ?

Is tri-sectional possible for sparse DF+ if I have  more than 3 fabrics?

Thanks



Best Regards,

Yuehua Wei

M: [phone] E: [e-mail]<mailto:[e-mail]>