|
|
|
|
|
by zamadatix
2030 days ago
|
|
Great intro article, saving this one for folks at work! For anyone confused why so many low buffer chips exist it depends greatly on the network needs. In some deployments going for deep buffers can actually make things slower overall. E.g. in enterprise healthcare I've really gravitated towards Trident (Trident 3 currently) level chips for most of the LAN/MAN and sometimes RAN as the feature sets are there and there isn't a strong driver for deep buffer per se just traffic priority for phones and a certain app once in a great while. I.e. it doesn't matter if you've got 16 GB of deep packet buffer built up, the telemetry packet needs to make it out the 10 Mbit pipe to the remote clinic now not at some point in the future. For WAN across regions or internet we move into some of those Juniper ASICs though and do rely on deeper buffer for the MANY concurrent sessions which are fighting understanding loss at higher latencies. Another interesting hardware rabbit hole to go down is the modular transceivers that go into these kinds of switches to provide different kinds of copper/optical connections instead of always being a fixed RJ-45 port like most would be used to at home. Crazy levels of engineering, especially at the higher end. I saw a really interesting presentation last year from Juniper on their use of silicon photonics for high density low cost 400 gigabit Ethernet and beyond (pdf warning) https://www.juniper.net/assets/us/en/local/pdf/nxtwork/silic... |
|