1

Environment:

  • Arista 7050sx3-48yc8 (two units)
  • HP Aruba 5412Rzl2 (backbone)
  • the backbone has an 8 port 10g module (two modules: 8x2=16 ports)
  • the sfps on arista and hp side are all 10g

We have an HP Aruba backbone with edge switches connecting to it. Recently we got two Arista switches to use for the new servers that are on their way.

We plan to connect the two Arista's (to each other) by using their 100G connection ports. So it will be a 8x100G connection between the two and act as a 86 port 25/10/1G switch.

We plan on connecting the Arista's to the HP backbone using a total of 2x8 (10g) ports. So in theory it will have a 160Gb throughput from the Arista's to the backbone.

What would be the best method for the Arista-Backbone connection? I'm not familiar with the unit that much. On the HP side i was thinking of making a "Trunk" for 16 ports. Not even sure if the 2 separate modules will allow it as a whole 16 port Trunk or if i would have to make them per-module (2x8).

What would the correct method be on Arista? Should I do LACP? Is there any better ways? I want to make the 16 ports act as a single 160g connection but also be able to run if any ports go down.

As I said my goal is to make the Aristas connect to the backbone with maximum throughput and have the ports be in a failover state (lets say a port goes down for some reason The remaining 15 ports should continue to work without any hiccups). The Aristas will hold our new servers that run vmware on them. So we want max performance and failover.

I am doing research but just need to be pointed to the right direction. Any advice will be appreciated.

Improve this question
0

1 Answer 1

Reset to default
1

The 5400zl2 is limited to 8-port LAG (LACP or static). You can use any ports for an aggregation group though, regardless of the physical module. I'd distribute the links to two modules.

Using Arista-side multi-chassis LAG, you could use a 2x4 LAG trunk (LACP preferred) to the 5412. Using two LAG groups of eight ports each to the 5412 would create a bridge loop, requiring xSTP to block one of the logical links (wasting its bandwidth).

If that bandwidth isn't sufficient, there aren't too many options - the 5400zl doesn't seem to support 25G ports (only 1G/10G/40G), so you're limited to 10G physical links and 8x10G per LAG.

[from comments] Note that an aggregation of e.g. 8x10G is not the same as a single 80G link - the aggregated link doesn't support any single flow faster than 10G due to the LAG traffic distribution scheme, avoiding out-of-order delivery. You'll only be able to utilize the total aggregation bandwidth when there's a number of flows significantly larger than the physical link number.

Traffic is balanced by the respective egress switch (server-> core by 7050; core -> server by 5412). The 5412 uses an L2/L3 MAC/IP address hash, so packets between any pair of IP end nodes always use the same port when leaving the 5412. I can't find that info for the 7050. (Arguably the best traffic distribution is achieved using an L2/L3/L4 MAC/IP/port hash, so multiple connections between any two hosts may use different ports.)

Improve this answer
4
  • Thank you for your response @Zac67. I have 16 ports available on both sides (Arista-Hp) for this purpose. Having a 2x4 connection would make it 80g if i understand it correctly. Also using 2x8 would provide 80g. So basically the latter would provide the same throughput but with a 80g failover. Right? If so I think using 2x8 would be a better choice for us because we have an abundant of ports. We already have the modules, sfps, cables and whatnot which won't be used for anything else.
    – shafuq
    Commented Oct 2, 2023 at 9:17
  • 2
    Basically yes. You should be aware that an aggregate link of e.g. 8x10G is not the same as a single 80G link - the former doesn't support any single flow faster than 10G due to the LAG traffic distribution scheme, avoiding out-of-order delivery. You'll only be able to utilize the total aggregation bandwidth when there's a number of flows significantly larger than the physical link number.
    – Zac67 ♦
    Commented Oct 2, 2023 at 9:25
  • Thanks again. So I won't be able to have a link speed over 10g no matter the amount of physical links within the LAG. There will be 6 rack servers connected to the Aristas. each with a 2x25g Ethernet connection. Now, with your information I understand that these servers will communicate with devices on the backbone using multiple 10g connection (8 to be exact). Each connection speed will be maxed out at 10g. BUT they will have multiple flows simultaneously. So the (throughput speed) sum could get up to 80g on any given time. Have I correctly understood how this design works?
    – shafuq
    Commented Oct 2, 2023 at 10:07
  • 1
    Yes exactly. Traffic is balanced by the respective egress switch (server-> core by 7050; core -> server by 5412). The 5412 uses L2/L3 MAC/IP address hashes, so packets between any pair of IP end nodes always use the same port when leaving the switch. Can't find that info for the 7050 though.
    – Zac67 ♦
    Commented Oct 2, 2023 at 11:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.