On 2013-12-28 22:18, Jakob Unterwurzacher wrote: > On 28.12.2013 12:18, rkwesk_ltsp wrote: >> I thought I had this figured out but I'd like to confirm: >> >> Configuration 1 >> >> unmanaged switch w 1 giga port and 16 100 ports not connected >> directly >> w router >> Server with two nics, one gigabit to giga port on switch and one 100 >> bit to router directly. >> Clients, mixed thin and fat but all with 100 bit nics to same switch >> to >> 100 bit ports. >> >> Configuration 2 >> >> unmanaged switch w 1 giga port and 16 100 ports, one of 100 bit >> ports >> connected w router >> Server with one nic (giga) to giga port on switch >> Clients, mixed thin and fat but all with 100 bit nics to same switch >> to >> 100 bit ports. >> >> In both cases (I think) there should be no overload and so the >> possible >> complications with flow control are avoided. >> >> Is this correct? > > No, I think you will experience head-of-line blocking if the switch > has > flow control enabled. > > Most of the traffic in LTSP is from the server to the clients. Say a > client watches an HD youtube video. This will easily generate more > than > 100Mb/s of traffic and will overload the client's 100Mb/s port.
As the switch's port to the client is also 100 Mps I think the client's 100mps nic cannot be overloaded per se. However, a fellow member of this list, alkisg, has since explained to me that the buffer on the switch will fill up since it is receiving packets from its giga port but can only release them at its 100 mbps port. It was this phenomenon (is it called buffer overrun?) that I hadn't thought about. This then forces the switch to send a pause frame to the server (in order for its buffer to be relieved.) Thus what you explain below here still happens. :(( > The > switch will send pause frames to the server, effectively throttling > down > it's 1Gb/s port to a slower speed. This will unnecessarily > affect all other clients' traffic, too - this is what is called > head-of-line blocking. Now I will show complete ignorance by asking whether if in fact the overload occurs even sooner (or more often) if an all giga switch were in place of the above described switch and the bottleneck was the 100 mbps nic rather than the 100 mbps port on the above switch? > That article explains it quite well (just multiply the speeds by a > factor of 10 ;) ) > > http://virtualthreads.blogspot.co.at/2006/02/beware-ethernet-flow-control.html > > > Jakob Thank you for the helpful link and for spending time with me on what seems so obvious to some of you. Richard ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk _____________________________________________________________________ Ltsp-discuss mailing list. To un-subscribe, or change prefs, goto: https://lists.sourceforge.net/lists/listinfo/ltsp-discuss For additional LTSP help, try #ltsp channel on irc.freenode.net
