[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Re (2): Linux hub

>---- Original Message ----
>From: bss@iguanasuicide.net
>To: debian-user@lists.debian.org
>Subject: Re: Re (2): Linux hub
>Date: Sun, 5 Sep 2010 14:56:53 -0500
>>In <[🔎] 380-22010905162433906@netptc.net>, owens@netptc.net wrote:
>>>>---- Original Message ----
>>>>From: peasthope@shaw.ca
>>>>>From:	"PT M." <pentie@gmail.com>
>>>>>Quoting from http://en.wikipedia.org/wiki/Network_switch,
>>>>>"Switches may operate at one or more OSI layers, including
>>>>>data link, network, or transport (i.e., end-to-end)."
>>>Lots of terminology confusion.  It used to be hubs were at level 1,
>>>switches at level 2, routers at level 3 and gateways above level 3.
>>That's not quite true; it's never been that clear cut, although some
>>like to believe that.
>>Hubs are "dumb" devices, almost always implemented entirely in
>hardware, and 
>>as such basically ignored everything above layer 1.  Mostly they
>should be 
>>avoided for TCP/IP over Ethernet, since they generate more traffic
>>required by Ethernet (which is layer 1 and 2).  In hubs, the most
>complex IC 
>>would not be complex enough to call a CPU in modern times.
>>Switches are smart devices, but not traditionally programmable. 
>They do use 
>>some RAM to store information about seen packets in order to make
>>about future packets.  They do not use out-of-band information to
>>switching decisions.  With TCP/IP over Ethernet, they generally
>operate at 
>>layer 2 (Ethernet, which is also layer 1) or 3.  Switches that do
>understand a 
>>layer above 2 will generally fall back to layer 2 when encountering
>a higher 
>>layer protocol they do not recognize.  Switches generally have a
>CPU, for 
>>handling the configuration interface, which might be extensive, even
>on layer 
>>2 switches, given all the parts of the Ethernet protocol.  However,
>>hardware is designed so that packets do not have to travel through
>the CPU 
>>Routers are programmable devices.  They have detailed configurations
>>static routes and routing rules.  In addition, they generally use
>>information (like BGP etc.) to update their routes and rules in
>>time.  They likely drop data that doesn't correspond to a layer 3
>>they understand, but could be configured to route it somewhere. 
>They will 
>>inspect layer 4 and possibly above data to aid their layer 3
>routing.  They 
>>will certainly have a CPU, but they also have switch-like hardware
>so that 
>>packets do not have to travel through the CPU bus.  Even packets
>that need to 
>>travel to the CPU for special processing may be routed but also
>saved so that 
>>traffic doesn't have to wait on the CPU; alternatively there may be 
>>specialized processors have more bandwidth then the CPU but much
>more limited 
>>A PC can pretend to be a router, switch, or even a hub.  However,
>unless it 
>>has some fairly specialized hardware, high load will reveal the PC. 
>The main 
>>cause of this is saturation of the bus between the network card and
>the CPU, 
>>since all the packets have to flow to the CPU and back.
>>Gateway is a much more generic term, but basically it connects your
>>with an outside network (or the Internet).  It will generally have
>at least 
>>minimal routing capabilities, but it may simply be an intermediary,
>linking a 
>>single switch/router to a single other gateway.  However, it may
>also be a 
>>feature-complete router and firewall along with doing layer 5-7
>inspection to 
>>enforce network policy.  For most personal and small business needs,
>it can 
>>simply be a PC; the traffic to outside networks being too
>constrained to need 
>>more bandwidth then is available across a modern PC bus, especially
>if all 
>>that bandwidth is through a single connection to a single ISP at any
>one time.  
>>Once external traffic reaches some level though, you'll need
>>hardware, so your gateway will likely be an "enterprise" router.
>>The specialized hardware that routers need, basically a
>"programmable" version 
>>of the hardware in a switch, was/is one of the main roadblocks on
>the way to a 
>>optical-only Internet backbone.  For the longest time, even when the
>>where optical cable, the routers where traditional electronic
>devices, so that 
>>the signal had to be decoded/encoded from optical to electric to
>optical which 
>>can introduce unnecessary delays.
>>TL;DR: That's an over-simplification or a case of nostalgia.
>>Boyd Stephen Smith Jr.                   ,= ,-_-. =.
>>bss@iguanasuicide.net                   ((_/)o o(\_))
>>ICQ: 514984 YM/AIM: DaTwinkDaddy         `-'(. .)`-'
>>http://iguanasuicide.net/                    \_/
I'm not disagreeing with you in practice but many years ago these
WERE the definitions the ITU and ISO dealt with.  IIRC it was the
vendors who screwed things up by introducing such products as
"swithcing hub"

Reply to: