Page MenuHomePhabricator

[cloudsw] enable 25G network
Closed, ResolvedPublic

Description

In order to allocate the newer high-density servers for ceph, we will need to enable the cloud network switches to be 25G at least on the ports for those new servers.

This task is to not forget and be able to track the work for that.

Related Objects

StatusSubtypeAssignedTask
Resolveddcaro
Resolvedtaavi
Resolvedcmooney

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
dcaro added a subtask: Restricted Task.May 8 2025, 9:17 AM
dcaro added a parent task: Unknown Object (Task).
dcaro removed a subtask: Restricted Task.
dcaro added a parent task: Restricted Task.

@ayounsi @cmooney feel free to use this task for this work, or link the one you are using for it, thanks!

@dcaro is there another task for those servers to be installed/provisioned?

We have support for 25G in Eqiad racks E4 and F4, and codfw B1. Not in eqiad C8/D5.

It's mainly just a matter of DC-Ops using the correct transceiver and allocating a compatible port. I think it's probably best dealt with in the provisioning tickets for the servers themselves than in a separate task.

@dcaro is there another task for those servers to be installed/provisioned?

It's linked as parent {T389851}, not yet bought, but we should keep this in mind

We have support for 25G in Eqiad racks E4 and F4, and codfw B1. Not in eqiad C8/D5.

We will need to support 25G on all of the racks, as we have to spread the nodes for high availability (specially critical if the hosts are that big)

It's mainly just a matter of DC-Ops using the correct transceiver and allocating a compatible port. I think it's probably best dealt with in the provisioning tickets for the servers themselves than in a separate task.

Sorry I found it. I added a note about C8/D5 in eqiad.

We will need to support 25G on all of the racks, as we have to spread the nodes for high availability (specially critical if the hosts are that big)

This is a blocker to be able to put those nodes in the cluster without losing capacity :/, we have to spread the nodes in different racks.

It maybe links with the high availability solution we were discussing, we have the budget this year to buy new switches specifically for HA right? Those will allow 25G iirc?

dcaro triaged this task as High priority.May 8 2025, 9:51 AM
dcaro added a subtask: Unknown Object (Task).

This is not a blocker anymore in general, will still be needed the more big hosts we get, but can wait for the general 25G everywhere some time.

cmooney claimed this task.

Closing this one @dcaro I believe all understand the current situation, and we can connect at 25G where it is possible. Please re-open if there is anything else we need to discuss.