Details
-
Type:
Story
-
Status: To Do
-
Resolution: Unresolved
-
Fix Version/s: None
-
Component/s: Science Platform
-
Labels:
-
Team:SQuaRE
Description
Fritz Mueller, Frossie Economou and Xiuqin Wu [X] we need to help inform Michelle Butler [X] on how to split up the physical resources between our two k8s clusters: production and development. The production cluster will be outside the firewall and will have NCSA exclusive kubctl. The development cluster will be inside the firewall with broader kubctl permissions. At least initially, there will be two deployments of the LSP with lsp-stable deployed on the production cluster and lsp-int deployed on the development cluster.
Cluster resources (I will update this description with actual numbers when I have them):
N nodes of type 1 (standard flavor)
M nodes of type 2 (dax flavor)
Note that the lsp-stable will need to support of order 100 users and the lsp-int will need to support of order 10 dev users.
Attachments
Issue Links
- mentioned in
-
Page Loading...
Activity
for SUIT needs, we can do the split as following:
lsp-int: one node with 124 GB ram; 40 cpu each with 10 cores (from the old SUIT nodes)
lsp-stabel: twos nodes with 124 GB ram; 40 cpu each with 10 cores (from the old SUIT nodes)
Frossie Economou is this question entirely about NCSA configuration and so can be closed?
For our part, we have resource estimates listed in https://sqr-018.lsst.io/. As currently deployed, a common per user resource load is 2 cores and 6MB RAM. Some demo notebooks require 4 cores and 12MB RAM per user.
I think that means we'd like the split to be 14 outside the firewall with two inside.