cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
443
Views
5
Helpful
2
Replies

HyperFlex Edge 3-Node-Cluster maximum Distance

Hello all,

i'm in discussions with a customer on a "to-be-sold" HyperFlex Edge 3-Node-Cluster, which should have at least a bit of "stretched"-flavour, due to customer's limited budget.

 

The customer asked, if it's possible to separate the servers and transfer the traffic via multiple switches, as seen in the following schematic view:

3_site_concept.png

So setup should be as follows:

  • 3 Firezones, 
    • Location A to Location B is about 30 meters cabling distance, 10G Multimode-fiber is planned
    • Location C to Location A & B is about 90 meters cabling distance, 10G Multimode-fiber is planned
    • each server is connected to the local single switch using 10G AOC-cables
  • Switches should all be Cat 930024T-E with 8x10G fiber module

 

I tried to find some more information about the maximum accepted RTT for the traffic within the HyperFlex system, but i understand of course, that this is not the normal case to set up the system. Customer has only a limited budget and going for a stretched cluster(what i of course also would prefer) is not an option. So if there is a technical limitation for this setup i'll think of a different architectural approach.

Theoretical i think, that it may work in the provided setup, but i know the switches cause some delay... So i'd appreciate to get some opinions or some comments, maybe some of you already built such a system  

 

Many thanks in advance!

Cheers,

Christian

2 Replies 2

Kirk J
Cisco Employee
Cisco Employee

Stretched clusters have a >5ms RTT so  you won't want to have any higher than 5ms between storage nodes storage data network interfaces.

What is proposed may 'work' but will likely provide suboptimal performance.

That design also brings in spanning tree complications as the storage data path has to traverse ports subject to spanning tree events that can briefly blackhole traffic.

With a normal dual switch design, the different 'fabric' sides involve two different switches, but the node to node communication is switched on the backplane of a single switch.  Also, the switch to switch trunks will have the storage data traffic competing with all other traffic.

IMO, while the customer may 'feel' like they have a more redundant solution proposed above, they are much more likely to have an outage with that design due to spanning tree convergence events, than simply putting the nodes connected to a couple of switches in the same room.

 

Kirk...

thank you for your thoughts and your input @Kirk J !

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community:

Review Cisco Networking products for a $25 gift card