Good morning all,
Hoping you can help answer a question surrounding a switch-fabric capacity alarm we saw today on a 7750 SR12 running with 3 active cards in slots 1,2 and 3 (IMM, IMM, ISM) and just 1 active CPM/SFM. The design includes a MC-Peer running the same which load share the traffic but using VPWS type services from host devices to either 7751 or 7752. The load share isn’t 50/50 for this reason as the load is “shared” based on the epipe configuration.
On 7751 we saw a switch-fabric capacity threshold alarm and all cards are showing as running 83% Switch-fabric capacity. Is the cause of this likely to be the CPM? How does this impact service? Is it just a slower transfer rate between ingress and egress and vice versa on this node?
Any help will be greatly appreciated.
Pip
Switch-Fabric Capacity
Re: Switch-Fabric Capacity
The alarm is generated when the when the total switch fabric capacity becomes less than the IOM/IMM capacity. The system is reporting that it is only possible to run the cards at 83% of full-duplex line-rate.
If you're getting the alarm for all the cards, it does sound like a CPM issue.
Depending on the linecard and CPM type, it may be expected that a single CPM does not provide enough switching capacity (for example, a 200Gb IMM and a single SF/CPM-4 will only provide 50% of slot bandwidth). But that is not likely to be the issue for a loss of 17% of capacity. The alarm has not always been present?
Not relevant for you, but there was a bug specific to the 7750 SR-a4/a8 which caused that alarm to be always generated.
If you're getting the alarm for all the cards, it does sound like a CPM issue.
Depending on the linecard and CPM type, it may be expected that a single CPM does not provide enough switching capacity (for example, a 200Gb IMM and a single SF/CPM-4 will only provide 50% of slot bandwidth). But that is not likely to be the issue for a loss of 17% of capacity. The alarm has not always been present?
Not relevant for you, but there was a bug specific to the 7750 SR-a4/a8 which caused that alarm to be always generated.
Re: Switch-Fabric Capacity
Thanks for the reply Mivens. Turns out we were able to rectify with “clear sfm 1” command. Looked like more of a firm chipset error as opposed to a hard chipset error.
Thanks for your response though, really appreciate it.
This error is attributed to a software defect, fixed in 14.0R4 (or greater) and does not require the HW to be replaced once the “clear” action is performed.
Regards,
Pip
Thanks for your response though, really appreciate it.
This error is attributed to a software defect, fixed in 14.0R4 (or greater) and does not require the HW to be replaced once the “clear” action is performed.
Regards,
Pip
Re: Switch-Fabric Capacity
Thanks for reporting back. Did you happen to also be given a bug reference (a DTS) for the that defect? As there doesn't seem to be anything in the list of fixed issues in the 14.0R4 release notes that obviously matches that (unless I've missed it).
