Date: Sat, 14 Dec 2002 19:23:43 -0800 (PST) From: Bsd Neophyte <bsdneophyte@yahoo.com> To: Greg 'groggy' Lehey <grog@FreeBSD.org> Cc: FreeBSD Questions <questions@FreeBSD.org> Subject: Re: Hubs and switches (was: uninformed qstn...) Message-ID: <20021215032343.64237.qmail@web20105.mail.yahoo.com> In-Reply-To: <20021215021454.GA53628@wantadilla.lemis.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--- Greg 'groggy' Lehey <grog@FreeBSD.org> wrote: > > BTW, I don't know if your switch is configured for cut-through > > switching or store-and-forward. If it is set for store-and-forward, > > I'd suggest you change it to cut-through. > > I don't care. We're looking at a difference of 200 µs here. well that's entirely your decision, however, if you have the equipment, might as well use it to it's fullest potential... otherwise, imho, there's no real point in having it. to each their own i suppose. > > I suppose if there isn't a big price difference between a generic > > switch and a hub, the switch would seem to be a better bet. > > There isn't, and that's the point we've all been making. here you go, from the horse's mouth... read and take what you want or educate yourself as you see fit. again, will reitterate that a switch is not always a better solution than a hub. ----- Sure, no problem. To be accurate, (and I'd agree it's a fine point), the point was that a switch is not always "better" than a hub ... not so much that there are cases when a hub would outperform a switch. I'm not sure if that point is clear. There are cases when a switch will perform at the level of a hub: specifically, when there are a number of ingress traffic streams all "aimed" at a specific egress stream (all ports, ingress and egress being the smae speed/bandwidth). At that point, the bandwidth is shared (on the egress stream) just as it would be for a hub (for example). The effect being like having a 5 lane highway choke down to one lane ... the traffic on the one lane is going to be "shared" by the cars on all five lanes. The reason I got into the whole discussion was that at the time (well, still .. to some degree) someone would ask "Can I use a hub to ..." , and the almost automatic reaction from most/many/all was "hubs suck, you need a switch, switches are better..." Well, in small networks (like, less than a couple dozen users), it doesn't really matter, especially if the network design is solid. The user's perception is that they are basically the same speed (given similar speeds, etc). Many corporations (including the one I work at) operated for YEARS with HUNDEREDS of users running on hubs (like the Bay 5000, for example). True, bandwidth demands have increased manyfold since, but good design (i.e., proper engineering) will win out over marginal design every time, even if the bad design has superior technology (or, at the very least, cost less for similar performance). There other issues depending on how sticky you want to get with it. For example, hubs will never result in "packet drop:" if a switch fabric is congested, the packet can time-out (and be dropped), in a hub, since the host can't send until it gets a clear channel, there is no (hub induced) time-out. A collision doesn't count as a transmission, the NIC (or stack) would re-transmit until sucessful (unless the retry count is exceeded, then it goes up the stack with an error, and the process up the stack tells it to initiate another attempt at transmitting). Switches induce more latency, even really fast switches. Switches (at a basic level) still propagate broadcasts and multicasts. Switches will still flood anytime it gets a MAC address it doesn't know. Now for the cop: Most of the time, switches will perform better than hubs. Full duplex and higher speeds is -usually- going to give the switch the advantage. As I mentioned (way up there at the top), I was kinda poking the knee-jerk "switches are better" respondents to see if they had any idea WHY switches are better (most didn't have a clue). It got 'em to think a little though, in most cases. The absolute best place for a switch is either a many-to-many (hosts-to-resources), where the circuit parallelism gives an apparent amplification of bandwidth, or, with lower bandwidth ingress being aggregated to a "fat pipe" on the egress side. The stats from the hub/switch test are still available at www.scottmac.net . We made a solid effort to keep the test fair across-the-board. Myself, Doug (L3Guy), and Dave (Mobly99) work (worked, actually ...we've been reassigned) in an Interoperability Lab and did this kind of testing (for hire, for profit) many times for our customers. The software and devices (Chariot and Smartbits) are industry-recognized "standards" for testing. Thanks for the interest, if I can clarify a point, let me know. Scott --------- well, that'll about conclude it from this end. just though i'd shed some light to dispell the "hubs suck" notion... because they don't always "suck." -Sameer __________________________________________________ Do you Yahoo!? Yahoo! Mail Plus - Powerful. Affordable. Sign up now. http://mailplus.yahoo.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021215032343.64237.qmail>