Because T346661: cloud: prepare codfw for expansion (racks, switches, ceph) we need to get 2 new cloudswitches in the codfw datacenter.
This task is to gather procurement details, like the particular hardware model, budget allocation, procurement time, etc.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | • aborrero | T326758 Research Openstack Deployment Paradigms | |||
| Declined | None | T342750 cloud: introduce a kubernetes undercloud to run openstack (via openstack-helm) | |||
| Unknown Object (Task) | |||||
| Resolved | Jhancock.wm | T342456 Q1:rack/setup/install cloudcontrol200[6-8]-dev, cloudnet200[7-8]-dev | |||
| Declined | None | T346661 cloud: prepare codfw for expansion (racks, switches, ceph) | |||
| Declined | cmooney | T346724 cloudswitch: codfw: figure out procurement |
hey @cmooney could you please advice on the cloudswitch models we would need in codfw to expand our capacity in that DCs?
Hi @aborrero
We should order the same as we already have for cloudsw1-b1-codfw. Which is Juniper QFX5120 (Broadcom Trident 3). To be precise, for each new rack we should order:
Capex:
| SKU | Desctription |
|---|---|
| QFX5120-48Y-AFI2 | Juniper 48x25G + 8x100G top-of-rack switch |
| S-QFX5K-C1-A2-P | Perpetual Advanced 2 licence for QFX5120-48Y Leaf |
I've spec'd the 'Advanced 2' license here. That supports EVPN/VXLAN, which at this stage would be my preference in terms of how to stretch the instances vlan between racks and provide automatic failover. It's a solution I would propose except the older QFX5100 devices in eqiad c8/d5 don't support it.
So, to match eqiad, we could go with just the 'Advanced 1' license (S-QFX5K-C1-A1-P, for BGP only). But it may be useful to have the ability to test this in codfw should we ever want to use EVPN to tackle the instances vlan.
Opex:
| SKU | Desctription |
|---|---|
| SVC-CP-QFX5-48YB | Juniper Care Core Plus Support QFX5120-48Y Leaf (12 Months) |
| SVC-COR-QFX5K-C1A2 | Juniper Care Core Support for S-QFX5K-C1-A2-P Leaf Software License (12 Months) |
In our conversation earlier the cloud team indicated they wish to investigate the use of Open vSwitch on cloud hosts to provide the required stretched L2 component for instances across racks.
This is the best way to support the use-case as far as netops are concerned, so there should be no need for us to look at solving the problem with EVPN on the switch side. We can therefore use the 'Advanced 1' license for the switches, making the updated BOM:
Capex:
| SKU | Desctription |
|---|---|
| QFX5120-48Y-AFI2 | Juniper 48x25G + 8x100G top-of-rack switch |
| S-QFX5K-C1-A1-P | Perpetual Advanced 1 software licence for QFX5120-48Y |
Opex:
| SKU | Desctription |
|---|---|
| SVC-CP-QFX5-48YB | Juniper Care Core Plus Support QFX5120-48Y Switch HW (12 Months) |
| SVC-COR-QFX5K-C1A1 | Juniper Care Core Support for S-QFX5K-C1-A1-P Switch SW (12 Months) |
I should also add that unless we are provisioning new racks, any rack allocated for this will already have a switch in it.
So we should first asses what racks are being made available, what hardware is already in them, and whether or not it needs to be replaced to support this project.
Basic rule of thumb, if we're talking about assigning existing codfw racks. Row A/B no new purchase required, row C/D are due for refresh in a year or two anyway so best to aquire new hw if using those.