Date: Fri, 14 Oct 2022 06:59:56 -0700 (PDT) From: "Rodney W. Grimes" <freebsd-rwg@gndrsh.dnsmgr.net> To: Wojciech Puchar <wojtek@puchar.net> Cc: freebsd-hackers@FreeBSD.org Subject: Re: is this FreeBSD problem or HP switch Message-ID: <202210141359.29EDxumV071505@gndrsh.dnsmgr.net> In-Reply-To: <586238fe-a9f-94b5-65b6-bc505551e9fb@puchar.net>
next in thread | previous in thread | raw e-mail | index | archive | help
> i have HP-2530-24G switch in home and 3 VLANs. > there was no traffic on 2 of them, one vlan have full bandwitch traffic > from computer A to computer B - at 1000Mbit/s > At this time ping from computer C to computer A (computer C have 100Mbit/s > card) was random between 1 and 1000ms. > > Just limiting artifically speed to 950Mbit/s solved the problem. > > Is this because switch behaves that way or can it be a FreeBSD problem > (all computers runs FreeBSD) If infact your traffic from A to B is at the line rate of the interface from the switch to B (A gigabit ethernet port) you are creating a situation known as head of line blocking. With a store and forward switch your packets from A to C are going to attempt to "tail insert" in the output buffer of the port connected to B, depending on how full that buffer is your going to see random ping times and/or drops. This is not a FreeBSD problem, different switches may behave differently, mostly dependent on the size of output buffers. Is the traffic from A to B something that has some type of congestion algorithm? If it is TCP can you turn on ECN, and enable ECN in the switch? -- Rod Grimes rgrimes@freebsd.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202210141359.29EDxumV071505>