Based on real world usage, currently a ratio of 288 compute nodes to 1
service node can be used as a general rule of thumb.
However, it is possible to scale beyond that with the right cluster
architecture design and tuning on the management node.
Questions to consider when trying to scale up:
- Do you want to boot all of your compute nodes disklessly simultaneously?
- If so, how large is your diskless image?
- Do you have a time requirement on how quickly the whole cluster must be
booted?
- Are there other reasons to add service nodes, for example, fault
tolerance?
- Where is the bottleneck in the service network when booting all of the
compute nodes and what is the best way to address it?
- Can you add a higher bandwidth ethernet adapter to the management
node?
- Can you add multiple ethernet adapters to the management node?
- Can you add service nodes?
From: "Vinícius Ferrão via xCAT-user"
<xcat-user@lists.sourceforge.net>
To: xCAT Users Mailing list <xcat-user@lists.sourceforge.net>
Cc: "Vinícius Ferrão" <fer...@versatushpc.com.br>
Date: 04/16/2021 03:56 PM
Subject: [EXTERNAL] [xcat-user] Service Nodes requirements and ratios
Hello,
I'm sorry to ask this repeated question in the mailing list, but I wasn't
able to find a proper answer with the DuckDuckGo-fu and looking in the mail
list archive.
What is the recommended ratio of compute nodes per service nodes and when I
should start considering it instead of leaving everything on the headnode?
Thanks,
V.
_______________________________________________
xCAT-user mailing list
xCAT-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xcat-user
_______________________________________________
xCAT-user mailing list
xCAT-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xcat-user
