gigabit pricing

Gigabit is too slow. Go for 10 Gb. ;-)

Fry's ad today has a gigabit NIC for $7.99, and five port switch for $19.99, no rebate required.

The brand is Airlink, which I don't know anything about, but that should give some idea where gigabit pricing is going.

-- glen

Geezz. That's all we need WinNICs. ;-)

Gigabit switches no longer have "uplink" ports, as all ports are now the same. So where older switches would often have 4 regular ports and 1 uplink, they're all available for regular connections now.

In article , glen herrmannsfeldt wrote: :Fry's ad today has a gigabit NIC for $7.99, and five port :switch for $19.99, no rebate required.

:The brand is Airlink, which I don't know anything about, :but that should give some idea where gigabit pricing :is going.

I wonder if the NIC is one of those ones that pushes nearly all the processing over to the CPU, like the old WinModems used to do?

5 ports on the switch is a bit odd, in that most gigabit chipsets handle either 1, 2, or 4 ports, and it would be the 4 port sets that would be least expensive. [See tomsnetworking.com from about March 2004 for some details on the chip wiring choices.]

But Yes, pricing is headed way down.

And in a way, that has me, in my role as a network administrator, quite concerned. With the prices of low-end unmanaged unconfigurable gigabit switches already having fallen to about the same price as low-end unmanaged unconfigurable 10/100 consumer-class switches, it is going to be very difficult for me to keep low-quality switches off of my network.

If one of our researchers needs a few gigabit ports for some project, it is going to be pretty much impossible for me to say, "No, you have to spend at least $C 5000 on a switch because I instinctively don't trust those off-brands" when the researcher can see "gigabit switches" advertised for less than $C 100, and when $C 5000 is enough to pay a PostDoc for a few months of work.

I had better have pretty strong arguments when the price difference is more than 50:1... especially when it's going to have to be me [whom the researchers don't have to pay out of their grants] that has to track down the problems rather than one of their people. You can predict the argument: "So you're telling me that whatever problems might -maybe- show up with this inexpensive switch, are going to take you two person-months worth of your labour to track down??" And of course the answer to that is "No". Now, if a cheap switch were to completely trash the LAN for a couple of hours then the work lost over all the employees would mount quickly, but complete trashing is rather unlikely: the most probable scenario is me ending up having to spend long unproductive hours to isolate some subtle problem, at perhaps a week's actual salary. The cost of mental aggrevation to me is not factored in. Yeah, like I need a plateload of frustrating network difficulties 'cuz I just don't have enough work to do :(

Now, who is making a Gigabit chipset that works that way?

With all due regard to tomsnetworking.com, there are several brands of inexpensive 5 port gigabit switch--if he says otherwise he's missed something.

I am always perplexed why people think that an 8-port switch is somehow more "natural" than 5, 6, or 10 ports. (The implication in the quote being that it is desirable to build 8-port switches.)

I understand that powers-of-2 are "naturally efficient" sizes for building *memory*, because the address space is log2 of the memory space. However, the number of ports on a communication device is not related to memory space, or powers-of-2.

In the early days of 10BASE-T, most hubs were built as multiples of 12 ports. There was a good reason for this, though. Cat-3 horizontal wiring was often implemented in 25-pair cables; this was the standard for pre-data phone wiring in offices. With 25 pairs, you can support 12

10BASE-T devices (at two pairs per device, with one spare pair). Rather than terminate the cable with 12 RJ-45 connectors, it was common to use a 50-pin "blue ribbon" connector; some hubs even came equipped with this as an option. Twelve-port hubs were therefore cheaper (to install), and fit naturally into the office wiring scheme.

Even today, 12-port (as well as multiples, such as 24 and 48) switches are common for 100/1000BASE-T, even though we no longer use 25-pair bundles when running Cat-5; the product marketing people just got used to building "12+1" and "24+2" port devices. They probably don't even know why.

The "naturally correct" number of ports for a switch should be logically related to the statistics of workgroup size. For example, if 95% of workgroups comprise 10 machines or fewer, then an 11-port switch (10 + uplink) would seem "natural". (I am not saying these numbers are correct; I am just showing the logic.) Of course, different port configurations will be more or less efficient in various application settings.

My point is just that there is no *technology* reason to build switches in multiples of 8 ports.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

Maybe you need to spend some up-front time qualifying unmanaged gigabit switches (or even 10/100 switches, I doubt most users can tell the difference), so you can say "yes, you can do it, but you must buy this particular switch".

Other than chip set configuruations or other hardware packaging issues. For example blocks of four RJ45 connectors are common. Other connector block sizes, such as 5 or 6 are less common. If you look at a 5 port gigabit switch, you'll see a block of 4 connectors and a single connector, even though they may be right next to each other.

Well, some do have more overhead than others. Some interrupt once per packet, others less often.

Five is convenient in that you can make eight ports with two chips, connecting one port of each together internally. (I don't know that they actually do that.)

Consider that you are at the end of a cable, and want to add more computers. A four port switch only allows two additional ports once you include the one to the existing computer and the uplink. So five is 1.5 times better in additional ports.

-- glen

Which is the "cause" and which is the "effect"? Do manufacturers make RJ-45 connectors in "4-packs" because equipment designers often build switches in 4-port, 8-port, 12-port configurations, or is it the other way around?

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

Actually it's X*(N-2)+2 where X is the number of chipsets and N is the number of ports per chipset.

"All" of my Gigabit switches are 5-port.

In article , wrote: : snipped-for-privacy@ibd.nrc-cnrc.gc.ca (Walter Roberson) wrote: :>And in a way, that has me, in my role as a network administrator, quite :>concerned. With the prices of low-end unmanaged unconfigurable gigabit :>switches already having fallen to about the same price as low-end :>unmanaged unconfigurable 10/100 consumer-class switches, it is going to :>be very difficult for me to keep low-quality switches off of my :>network.

:Maybe you need to spend some up-front time qualifying unmanaged :gigabit switches (or even 10/100 switches, I doubt most users can tell :the difference), so you can say "yes, you can do it, but you must buy :this particular switch".

I thought of that, but I don't have a lot of spare budget to buy switches just to test out, and I don't have a lot of spare time or equipment to set up meaningful testbeds. My job would in a way be easier if we were business rather than research, in that I would then be able to use the "We can't afford to have the network down, so we

-need- to spent the money and my time to do this properly" approach. I am under serious pressure to spend -less- time on networks, not more.

I suggested to all the LAN and security admins across our organization that we pool together our experiences with gig switches, but I only got back one "count me in" and one useful annecdote. Others either didn't anticipate adding any switches to their network for several years yet, or else were content to follow the recommendations of our central WAN networking group whose focus is on providing connectivity to our HQ campus: that group can easily put together a business case for needing high end equipment for the core links at HQ, but we in the regional sites cannot afford even the maintenance contracts on the class of equipment that HQ needs.

In article , Rich Seifert wrote: :I am always perplexed why people think that an 8-port switch is somehow :more "natural" than 5, 6, or 10 ports. (The implication in the quote :being that it is desirable to build 8-port switches.)

:I understand that powers-of-2 are "naturally efficient" sizes for :building *memory*, because the address space is log2 of the memory :space. However, the number of ports on a communication device is not :related to memory space, or powers-of-2.

No, but the number of ports is naturally multiples of N, where N is the number of ports per chipset. Gigabit chipset designers happen to be working at 1, 2, or 4 ports per chipset. I don't have any experience with chipset design to know whether there are advantages to even numbers or powers of 2 at the chipset level.

:Gigabit switches no longer have "uplink" ports, as all ports are now the :same. So where older switches would often have 4 regular ports and 1 :uplink, they're all available for regular connections now.

I have looked at a small number of gigabit switches in a bit more detail, and I have found that asymmetric service is still common. The ports might all have the same nominal bitrate, but the internal buffering for the ports are not always the same.

On the lower end devices (and not so low end either!), I have found more than one device that cannot handle wire rate simultaneously across groups of its ports [shared processing in a chipset that handles a group of ports], but which can handle higher rates on a small number of its ports (one or two). This is most obvious (most readily findable in the documentation) when one of the ports is a GBIC or SFP and the others are autonegotiating RJ45 ports.

The economics of this make some sense: the devices are aimed at workgroups of hosts which have fairly idle traffic a lot of the time (e.g., Windows desktop boxes) with large transfers relatively infrequent and not often happening simultaneously with the different hosts. Typical underprovisioning logic. The port or two that are configured for higher rates would be intended as the uplink, not in the sense of having a higher bitrate, but in the sense of being intended to carry a higher average traffic load.

Most of them in stock starting at under $50.00.

It is clear that _somebody_ is making a bargain basement 5-port gigabit chipset.

(snip)

I think eight is a good size for a small box. The 24 and 48 port are rack mount size, but eight would be a waste of rack space. Also, if one is comparison shopping it is hard to compare if all companies make different numbers or ports.

(snip)

Also, workgroup size changes fast enough that you want some extra ports. Five might be a good size for a single small office, and eight for the next increment from five.

-- glen

Yes. ;-)

Actually, it's hard to say. The switches and hubs I've seen, tend to be multiples of 4, except those 5 port ones. Is it due to chip design? I read an article in Linux Journal, a few years back. I seem to remember the chip mentioned in the article, had 4 ports and a "bus" interface, which could also be used as a port. I don't know the history of these devices, but I wouldn't be surprised, if there was something that at one point made multiples of 4 a good choice. The industry is full of this sort of thing. For example, in the telcom world, a channel bank (used to connect voice circuits to a T1) has 24 channels (in North America), so everything that's designed to work with it, has 24 voice circuits. Then when you get to DS1-DS3 multiplexors, the magic number is 28. Again, all the patch panels etc., will have 28 circuits. I don't know how those numbers were originally derived, but I bet it made sense to some engineer at that time, possibly due to the use of some other device. In another message, you mentioned 25 pair cable for phone systems. Prior to those channel banks (mid 60's) 25 pair cables were not so common, as the older analog systems, came with various numbers of channels.

Now if only they made a GIG version....