To Scott Leibrand's point, the potential to discover MTU in-band does exist - it's codified in RFC 4821. Linux implements it via the tcp_mtu_probing sysctl; Windows servers call it "PMTU Blackhole Router Detection".
I think that calling this presentation "The Case Against Jumbo Frames" and then talking about flaws in MTU path discovery and lack of standarization is not very smart. Jumbo frames give you +20% speed on the same hardware. Who wouldn't want that?
I got a throughput hike of 4x after enabling jumbo frames. For backup and restores of data in TBs , we need Jumbo frames. Offcourse, we can avoid router and keep traffic till switch level...
My ISCSI transfer speeds to and from my NAS increased significantly after enabling jumbo frames on my small internal network and I would never go back!!!
@@nerdspark8084 He is being very misunderstood. He is not against solving with packet window size and frames issue. He is saying that turning on jumbo frames eg: setting MTU to 9000 isn't a solution, cause in a few years, we will have the same issue!(actually, we are already there, 9000 MTU still has a s..tload of overhead on 10, 15, 40+ gbe) We need to get rid of MTU size all together, and come up with a way to eliminate this flaw in ethernet negotiation.
watched 3 minutes of this. my brain is fried everytime i hear him say jumbo packet as in a layer 3 technology and not jumbo frame which is on layer 2. there is no such thing as a jumbo packet. half the coments here also say jumbo packet. packets go inside frames multiple 1500 sized packets can be inside a 9000 jumbo frame
This guy just calmly destroyed my weekend
To Scott Leibrand's point, the potential to discover MTU in-band does exist - it's codified in RFC 4821. Linux implements it via the tcp_mtu_probing sysctl; Windows servers call it "PMTU Blackhole Router Detection".
So all the people on YT telling me to run 9000 Jumbo frames to speed up my 10Gb/s LAN are incorrect????
I think that calling this presentation "The Case Against Jumbo Frames" and then talking about flaws in MTU path discovery and lack of standarization is not very smart. Jumbo frames give you +20% speed on the same hardware. Who wouldn't want that?
I got a throughput hike of 4x after enabling jumbo frames. For backup and restores of data in TBs , we need Jumbo frames. Offcourse, we can avoid router and keep traffic till switch level...
My ISCSI transfer speeds to and from my NAS increased significantly after enabling jumbo frames on my small internal network and I would never go back!!!
@@nerdspark8084 He is being very misunderstood. He is not against solving with packet window size and frames issue. He is saying that turning on jumbo frames eg: setting MTU to 9000 isn't a solution, cause in a few years, we will have the same issue!(actually, we are already there, 9000 MTU still has a s..tload of overhead on 10, 15, 40+ gbe) We need to get rid of MTU size all together, and come up with a way to eliminate this flaw in ethernet negotiation.
watched 3 minutes of this. my brain is fried everytime i hear him say jumbo packet as in a layer 3 technology and not jumbo frame which is on layer 2. there is no such thing as a jumbo packet. half the coments here also say jumbo packet. packets go inside frames multiple 1500 sized packets can be inside a 9000 jumbo frame
I have news for you; no Ethernet spec in the world carries multiple IP payloads in one frame, barring encapsulation.
Thanks for putting me to sleep.
Hey! Cmon be nice.