[GTALUG] I’m obviously way behind in my reading: IBM owns Redhat

[Alvin Starr]

On 5/28/20 9:10 AM, James Knott via talk wrote:
> On 2020-05-28 08:30 AM, Alvin Starr via talk wrote:
>> Generally speaking fiber is still more expensive than copper wiring 
>> but the difference is not nearly as bad as it once was.
>> The biggest thing is the price of CAT-5,6,7... is way down the price 
>> curve due to volume and the fact that most of us can afford the 
>> hardware to terminate copper cable.
>> The cost of fiber termination is still way up there. 
>
> The cost of fibre patch cords isn't that far from copper:
> https://www.fs.com/c/fiber-optic-cables-209?gclid=Cj0KCQjwwr32BRD4ARIsAAJNf_1q8st39oVuCo89F6NQOX8dAx0yCm8-isqGkMAxY3_IgfIIJgQsaWsaAm7cEALw_wcB 
>
>
FS.com is a good source of inexpensive patch cables and networking products.
But still a 1M OM3 cable is ~$5 and a 1M CAT5e is ~$2.5.
Not a huge difference but if you need a thousand of them it starts to 
add up.
I am sure a Cisco SFP is more than $10 from Cisco

If your doing the cabling then the cost of the termination and testing 
equipment  can become a stumbling block.\
There is also the cable prep and cleaning that is needed with optical 
fiber which is not part of the CATx

> SFP modules will set you back less than $10.  Compare that to the cost 
> of a Cisco switch or router and you'll find the difference is trivial.
>
> There's also the advantages such as immunity from electrical 
> interference, shock hazard, no longer limited to 100M (in fact, it's 
> possible to cross the Atlantic, without repeaters). Unbelievable 
> bandwidth (theoretical maximum 2.5 petabit, IIRC), less volume 
> (important with larger cables) and more.
Absolutely.
There are some real advantages to fiber and I did not say that fiber did 
not have its advantages just that it is still a little more expensive 
than CATx.

>
>> Redundant networks over the same cable(bundle) is not really redundant.
>
> Who said anything about being in the same bundle?
True. That was my bad.
I made an assumption from your comment about the cable being along the 
LRT right of way.
But that being said all cable along a single right of way is in essence 
a single point of failure( think back hoe ).
Multiple fiber bundles does not make that problem go away.

>
>> Ethernet is not a real-time protocol.
>> There are a number of ways to cover that shortcoming  but Ethernet 
>> networks will never have deterministic throughput.
>> For most things this is OK but if you need microsecond level timing 
>> then your looking at some other protocol.
>
> Token ring was deterministic, in that there was a maximum time, before 
> a device could transmit.  This compared with half duplex Ethernet 
> (coax or hub), where there's a random factor.  However, that 
> difference doesn't exist with switches, where you can assign priority, 
> scheduling and more.  The limiting factor with Ethernet is something 
> called "funneling", where packets from multiple sources are all 
> heading to the same port, which will cause a backup.  VLAN tags can be 
> used to provide priority, as well as ToS in IP.  So, it all boils down 
> to proper engineering of the network.
>
Still Ethernet is not real-time.
There are a lot of things that can be done to make it more amenable to 
human scale real-time but hard real-time operation is just not part of 
the mix.
Once you hit a switch it gets worse.
There may be some switches out there with deterministic queuing but I 
don't know of them.
>> Throwing out real-time response and using CSMA/CD half duplex 
>> operation made the initial Ethernet hardware much cheaper than its 
>> competitors.
>
> When's the last time you saw half duplex Ethernet?  Also, most of that 
> gear ran at 10 Mb, with a small amount capable of 100 Mb. However, 
> switches killed half duplex.  Of course, 100 Mb was left in the dust 
> years ago, with 100 Gb, or more, now used in many applications.
> https://en.wikipedia.org/wiki/Terabit_Ethernet
Yesterday.
I still have some thinnet hardware in my junk pile.

The design choices early on have effects on the uptake of a technology.
Cheap Ethernet cards made for more people using the technology and more 
people using the technology leads to more companies developing hardware 
to extend the technology.
Rinse and repeat.

The openness or propritaryness  of the protocols may have also played 
into those cost calculations.

I remember a time when it looked like ATM might take take a chunk out of 
the Ethernet installed base but the cards never came down in price 
enough to make it happen.

I did have a friend who made a lot of money by selling his ATM card 
startup to Cisco.
Losing a manufacturer may have help in the demise of cheap ATM cards.

>
>> As for IP. It is best described as resilient and not redundant.
>> The protocols on top of IP make for a very robust network that will 
>> generally route around failures.
>> But they are even further from deterministic.
>
> IP and Ethernet are best effort.  Then again, so was token ring. They 
> all rely on some end to end protocol, such as TCP to provide 
> protection against data loss.  You'd have to go back to X.25 (yes, 
> I've worked with that too) to get such protection on a per hop basis.
>
> As I mentioned, almost my entire career, going back to 1972 has been 
> in telecom, computers and networks, so I do have some idea about what 
> I'm talking about.

I was there also and around the same time along with dealing with 
aircraft avionics where there are some hard real-time and redundancy 
requirements.
I also know a little about this.

I did not suggest that you did not know what you were talking about I 
was just pointing out some subtler points.

-- 
Alvin Starr                   ||   land:  (647)478-6285
Netvel Inc.                   ||   Cell:  (416)806-0133
alvin at netvel.net              ||