Administered by:
I still regularly think about the CDN customer I was working with who was using a 7060SX2-48YC6 for one of their POPs.
About two dozen web servers, all connected to the switch with 25G NICs, and then the switch was carrying a default route out transit and had a 10G peering port on an Internet Exchange.
They had two dozen 25G servers all trying to push traffic out a single 10G port on a switch platform that effectively only had 5.5MB of packet buffer for the speed change.
Needless to say, this was about the textbook application where shallow buffer switches are going to give you problems. #networking
@kwf Eeyupp....
They should've properly sized the internal connect to avoid such bottlenecking...
@kkarhan Needless to say, they didn't want to spend the money on upgrading the IXP port either. They just wanted me to explain why they were dropping packets all over the place even though SNMP said they were moving less than 1Gbps on the IXP
@kkarhan@mstdn.social@kwf that reminds me of news stories that #DTAG abusing #Cogent & #Level3 as #Upstreams and refusing to upgrade the bandwiths, causing #PacketLoss and instable connectivity out of their #ego not wanting to peer at #DECIX and have properly sized ports.
This even went into lawsuit filings, but propably got pulled due to arbitration.
After all, I remember colleagues saying one should not exceed 30-50% of the network capacity prior to upgrading just to have enough breathing room.