[kwlug-disc] CAT6 - worthwhile?
Cedric Puddy
cedric at ccjclearline.com
Sun Nov 24 20:46:47 EST 2013
Hi All,
A bunch of thoughts re: switches, Cat5e/Cat6, and so on. Hope it helps.
Cat5e / Cat6 is nothing to do with fire ratings -- here's the Cat6 wikipedia page: http://en.wikipedia.org/wiki/Category_6_cable
Regarding Fire rating of cables, see this wikipedia article: http://en.wikipedia.org/wiki/Plenum_cable (basically "plenum rated" cable has fire retardants, and fire codes require it to be used in "air handling spaces"; "riser cable", so called because it's primary use is to go between wiring closets in dedicated ducts, does not have the retardants. Likewise, patch cords and whatnot typically do not have retardants because the fire retardant plastics are stiff and easily damaged, which is bad for cables at your desk.)
Regarding choosing switches...
It's not necessarily bad for the sua switches backplane bandwidth to be less than the product of it it's ports -- it's a definitely a something that changes the prices of a switch.
More expensive switches can handle their full aggregate bandwidth, yes, but:
1) they likely have loud fans (not great if being installed in a home office, or in an employees cubicle
2) you may simply *not need* full wire speed on every port -- take the example below, where the bulk of your traffic is concentrated on a given port:
2a) if people are mostly accessing a server, then the most they'll be able to consume is the bandwidth of a single port
2b) if people are mostly accessing the internet, then they'll not be able to use much more than the bandwidth of the incoming internet feed
2c) you might still want a GigE switch to keep packet latencies low, enjoy the benefits of a faster signalling rate for individual nodes
3) you may be on a budget, were money is as important a factor to balance as performance
4) you aren't going to find *any* switches on the market that cannot do full wire-speed on at least... oh... probably half their ports simultaneously. Actually, I just went looking on www.procurve.com (HP's switch site), and can't find any switches that don't have more fabric performance than they do ports, and lots of switches that have a multiple. (Eg: 8 ports, 10 Gbps fabric; 24 ports, 128 Gbps fabric; 48 ports, 176 Gbps fabric... and so on.)
What with the return to dumb terminals and centralized computing resources, it's becoming less and less necessary to have high performance networking to the individual computers in the network. Combine that with the rise of VoIP, 802.11af, and the fact that most phones do PoE and have only 100 Mbps ports you get the following result:
a) clients buy thin clients, which only need a small fraction of 100 Mbps performance (say, 0.4-4 Mbps, depending)
b) clients rollout PoE switches (probably GigE, but just as easily 100 Mbps)
c) clients put a PoE VoIP handset on the desk, daisy change the thin client off of that (eg: 100 Mbps networking "to the desktop")
d) server(s) connect to the switch via GigE uplink ports, if it's a 10/100 switch (most 10/100 these days switches have 4)
e) wireless with WPA2 (hopefully with Enterprise rather than PSK) looks after the laptops and tablets.
f) Yes, it's nice if the switch supports LLDP-MED to make it trivial to separate out phone traffic from data traffic (the idea is that phones don't have a lot of horsepower to spend examining packets, so if you have phones directly on a busy network, they can get bogged down looking at packet headers, 99.8% of which aren't of interest to them in the first place, so better to put them on a nice quite VLAN).
Tada. The client has a "fast enough" network, needs only a fraction of what Cat5e signalling provides, doesn't care much least about raw switch performance, needs only one port per desk for the most part, and so forth.
I'm seeing the above fairly routinely in my client base, at least.
Now, when you start designing the network for the *servers* -- say, hypervisor hosts connecting to a iSCSI SAN, multipathing, various of VLANs, future-proofing, you care a lot less about power consumption/noise/cost of the switch, and a lot more concerned with feature set, performance (because now you have lots of ports that could need to run at wire speed), whether cutting a corner now might screw you over in some expensive way later, etc.
Personally, I don't get the need for Cat6 in a lot of cases. For long runs, Cat6 will help assure you that you can do error-free data transmission (eg: it might be technically within the range of Cat5e, but Cat6 will give you a lot more margin for installation error!)
The trends I'm seeing suggest that it's going to be a very long time before Cat5e has run it's course. Naturally, if it's a job that would be very hard to do again later (wiring a residence during construction, a building that you cannot make a good prediction around future use of the cable plant), a job with any very long cable runs, etc, then by all means, go Cat6.
If it's a small job, easy to do again later because it's all open, where it's an infrastructure that's not going to be required to do anything crazy, then there may well be every reason to go Cat5e, and do whatever you like with the savings :)
As discussed in Wikipedia and elsewhere, the difference between Cat5e and Cat6 is signalling bandwidth. Bandwidth translates either maintaining integrity while throwing a slow signal a long way (eg: 100 Mbps ethernet over a long distance on Cat6), or ability to do a high speed signal over a short distance (eg: 10 GigE over RJ45 copper*). Cat6 (and Cat5e) are *systems* that are a combination several things:
- precisely how the cable is constructed (rate of twist on the pairs and groups of pairs, composition of the conductors, choice of dielectrics),
- how the cable is run (has it been stepped on? kinked? run the cable parallel within 6" of AC power cables? wrapped it around a transformer a few times?)
- exactly how the jacks at the ends of the cable are constructed,
- and how good a job the cable installer does of installing the jacks on the cable (has he nicked a conductor or two? maintained twist up to 1/4" of the insertion point on every pair? punched each conductor into the jack evenly and fully? yanked on it by accident after punching it down? accidentally left an end of a conductor extending past the insertion point? etc..).
- choice of patch cables (are they pre-manufacturered Cat6 patch cables? are they at least 3 ft long?)
Also, you can hand build Cat5e patch cables that meet spec (Why you would try, I don't know! But some people can manage), but humans *cannot*, I am given to understand, by hand achieve the precision needed to build Cat6 patch cables that meet spec. (It's been more than 10 years since I've had to build a patch cable, and if I found any of the techs I work with to try making one as anything other than a joke or experiment, I'd probably opt to never work with that tech again. IMHO, There Is Simply No Reason to make patch cables by hand.).
All the Best,
-Cedric
* RE: 10 GigE over RJ45 copper: Never do this. Yes, you can buy 10GigE network cards that have RJ45 plugs. But I think they should be avoided. Because:
1) all the RJ45 10GigE cards I know of use use 2-3x the power of a CX4 or fibre connection
2) all the RJ45 10GigE cards I know of have several times the latency of a CX4 or fibre connection
CX4 is a external connection style that is used for External SAS, 10 GigE over copper, FC, PCI-E expansion busses, and Inifiband. It's a multi-lane serial connection. Fairly heavy cable, but has a nice locking connector, same power consumption and latency of optical fibre without the routing and maintenance complexity of optical fibre.
Fibre can be bad because it can be tough to maintain to spec -- you've all seen pictures of rat's nest cabling in data centre racks. Unless you invest the $1000-2000 extra, per rack, to get all the fancy fibre race-ways, bend radius pieces, separate cable managers for power/copper wire, etc, or are the most anal person *EVER* about tidy wiring, then your rack full of optical fibre will ultimately end up getting tangled, have fibres exceeding there bend radius tolerances, potentially have downtime because you snapped a fibre, etc. Copper is far more forgiving. Also, connections can be easily degraded by dirt n' dust, and hard to clean if the bottom of the female side of the connection gets dirty. For this reason, I think fibre running around in homes, most small business racks, etc, is not a great idea. In a pristine server room environment, maintained by pros, with a powerful incentive to do whatever it takes to keep it all looking like it's sitting on the showroom floor, by all means, I say fill your boots.
On 2013-11-24, at 10:10 AM, John Johnson <jvj at golden.net> wrote:
> Re: "each device on copper has a full speed dedicated connection"
>
> AFAIK not necessarily
> While with point-to-point copper there is no "ether" as there is with wifi the hardware internals in a router or switch may include a common bus. i.e. shared media inside the box.
>
> I doubt that two servers (A and B) streaming data to two workstations (C and D) as in A to C and B to D at the same time through a common router / switch would each enjoy the maximum throughput that would be available if only one server was streaming.
>
> AFAIK CATx ratings were primarily the fire ratings as opposed to data throughput. (Wikipedia here I come).
> And the toxicity mentioned earlier is due to materials designed to suppress fire traveling along a cable from a flame source to another area in a building.
>
> John VanO and others may be able to clarify.
>
> Apologies: Paul (2013-10-26) says the entire thread[s] is[are] worth reading. Maybe I should do so - before my next comment.
>
> JohnJ
>
> On 2013-10-26 17:40, unsolicited wrote:
>> No. To expect it is unreasonable.
>>
>> Each device on wi-fi halves the throughput for each device on the wi-fi. So your 300Mbps, which you will never achieve (below), goes to 150 as soon as you put something else on. Like a printer. Even just a device turned on has a hearbeat / continuously scans for connectivity and takes up some of the time slices.
>>
>> And wi-fi (let alone usb over esata) has much more protocol overhead such that data throughput is seriously degraded. Add in checksumming and encryption and it gets even worse.
>>
>> - each device on copper has a full speed dedicated connection. (Doesn't mean two devices on a switch trying to talk to the next hop beyond the switch get it - even gigabit switch to switch is still only gigabit. BUT - each device will get full speed when it's got the wire, and each device will get through what it needs to do faster. If the two devices are talking to each other, then they talk full speed to each other, and something else can talk full speed up the line to the next switch - simultaneously. In such a scenario, each device would be doing much less better than 1/4 full speed, wifi.)
>>
>> 300Mbps is only achievable through dual-band *5GHz* - not always achievable with other networks around.
>>
>> Just as I noted its ludicrous that not everything comes with Gbps NICs these days, even fewer come with 5GHz, let alone dual, wi-fi radios.
>>
>> So, sorry to say, your shiny new 5GHz router, connected to your 2.4GHz single band printer has bought you ... nothing.
>>
>> Even throwing down a network cable to the printer, at 10Mbps, will gain you a 15x speed increase, IIRC. And every other wi-fi device will gain from not competing with the printer any more.
>>
>> - just like turning off wi-fi on your phone will do the same.
>>
>> I ran speed tests when I got the last laptop - maybe it's useful. http://kwlug.org/node/750
>>
>> In the end, the only thing that matters is - is it fast enough for you?
>>
>> In the end, wi-fi is a fantasy. Do backflips to figure out copper if you have to - it's worth it. 'cause there will always be another wi-fi device coming along to chew up what wi-fi bandwidth you have - like your phone (no copper connection equivalent), or someone else's, needing the internet ... 'just for a moment.'
>>
>>
>> On 13-10-26 11:26 AM, John Kerr wrote:
>>> My current router at work is rated at 56mb. I just purchased a new router
>>> rated up to 300mbps. Great except the Internet comes in at 25mbps.
>>>
>>> I have purchased a new wireless printer as well.
>>> So I bought the new router so that any printing jobs will be done faster.
>>> Am I in my right mind on this?
>>>
>>> To me it is the last 50 feet that are important. Will there soon be a
>>> wireless Internet service that will give us the speed that our routers are
>>> capable of?
>>>
>>> Cheers
>>>
>>> John
>>>
>>>
>>>
>>>
>>> On Sat, Oct 26, 2013 at 2:30 AM, unsolicited<unsolicited at swiz.ca> wrote:
>>>
>>>> OK, but my real basic question is, to what end?
>>>>
>>>> Gigabit has even now not permeated enough of the world. (Why any laptop
>>>> still comes with 10/100 is beyond me). Even my USB 3.0 / gigabit adapter
>>>> can't saturate the gigabit.
>>>>
>>>> Future proof for what (copper wise)?
>>>>
>>>> If the world is going tablets and phones - that's wifi, not copper. Even
>>>> if you have copper and an AP at each room for wi-fi devices to connect to,
>>>> no amount of devices on that wi-fi will ever saturate the gigabit - wi-fi
>>>> will never be that fast. (?)
>>>>
>>>> Home wise, I'm not prepared to even put out for multi-run bonding - the
>>>> equipment required at each end is extraordinarily expensive (for home). I
>>>> don't imagine it's any different for 10Gps CAT6 ethernet, let alone fibre.
>>>> And if it's fibre we get to, the copper run, 5e or 6, isn't going to be
>>>> useful.
>>>>
>>>> So if most things can't saturate gigabit now, and fibre is going to need
>>>> another run anyways if we get there ... future proof for what (sorts of
>>>> beasties / media)?
>>>>
>>>> I'm not objecting to 6 over 5e, I just wonder ... for what?
>>>> - especially given the more expensive equipment required at each switch
>>>> point, and the tighter bend and untwist limits for 6. I'd bet every home 6
>>>> installation breaks at each jack / switch / 5e device<->jack cable.
>>>>
>>>> If you've bent a cable, what, more than 30 degrees, or untwisted a pair,
>>>> or untwisted pairs more than 1/2 inch - you've just made using cat 6
>>>> pointless.
>>>>
>>>>
>>>> So my real question was ... what's coming that might need 6 over 5e?
>>>>
>>>> In house HD video distribution?
>>>>
>>>>
>>>> On 13-10-25 04:57 PM, John Van Ostrand wrote:
>>>>
>>>>> Personally I think any new installation should use Cat 6. It's only
>>>>> marginally more expensive than 5e but could future-proof your house a
>>>>> little more. That said 5e will perform very well in a house since runs
>>>>> tend
>>>>> to be short and will work in cases where Cat 6 is supposedly required. The
>>>>> way I look at it is that the time spend installing is the the most
>>>>> expensive cost (even when done yourself) so using a higher grade cable
>>>>> future-proofs so you can avoid pulling everything out and re-doing cable.
>>>>> Sometimes I'll use 5e jacks because those are easier to replace.
>>>>>
>>>> .
>>>> .
>>>> .
>>>>
>>>>
>>>> ______________________________**_________________
>>>> kwlug-disc mailing list
>>>> kwlug-disc at kwlug.org
>>>> http://kwlug.org/mailman/**listinfo/kwlug-disc_kwlug.org<http://kwlug.org/mailman/listinfo/kwlug-disc_kwlug.org>
>>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> kwlug-disc mailing list
>>> kwlug-disc at kwlug.org
>>> http://kwlug.org/mailman/listinfo/kwlug-disc_kwlug.org
>>
>>
>> _______________________________________________
>> kwlug-disc mailing list
>> kwlug-disc at kwlug.org
>> http://kwlug.org/mailman/listinfo/kwlug-disc_kwlug.org
>>
>>
>> -----
>>
>> Checked by AVG - www.avg.com
>> Version: 2012.0.2242 / Virus Database: 3222/6282 - Release Date: 10/25/13
>>
>>
>
>
>
>
> _______________________________________________
> kwlug-disc mailing list
> kwlug-disc at kwlug.org
> http://kwlug.org/mailman/listinfo/kwlug-disc_kwlug.org
| CCj/ClearLine - Hosting and TCP/IP Network Services since 1997
| 118 Louisa Street, Kitchener, Ontario, N2H 5M3, 519-489-0478x102
\________________________________________________________
Cedric Puddy, IS Director cedric at ccjclearline.com
More information about the kwlug-disc
mailing list