Practicalities of subnets

Hi guys,

I'm new to networking and have some questions on my mind that I hope
can be answered.

I understand that subnets make managing networks easier, but what
benefits are there of having hosts in the same subnet as opposed to
different subnets? why are some servers placed in the same subnets and
some placed in different subnets when they all need to communicate with
one another?

Thanks a million

taihc wrote:
> Hi guys,
>
> I'm new to networking and have some questions on my mind that I hope
> can be answered.
>
> I understand that subnets make managing networks easier, but what
> benefits are there of having hosts in the same subnet as opposed to
> different subnets? why are some servers placed in the same subnets and
> some placed in different subnets when they all need to communicate with
> one another?
>
> Thanks a million
>
That opens the door for a lot of possibilities, but here are a few
things that might help to answer your question.

1) Most networks have Windows to deal with. Now I'm not saying that's a
good or a bad thing, just that it's so. Windows boxes use broadcast
based announcements on top of ARP and the usual IP protocol babble that
all networks have. So breaking networks up into subnets (physically or
virtually) contains broadcasts because routers don't pass broadcasts.

2) VLANs are used to segregate networks at the layer-2 level. In order
to pass traffic between VLANs a router is required. Since routers route
between different networks (subnets), each VLAN must be in a different
subnet.

The primary benefit to having hosts in the same subnet is that they can
reach other via broadcast. Again, on a Windows/Samba network, that means
"My Network Places" will work without a WINS server. Servers are
generally placed in the subnet with the hosts that will access it the
most. The primary reason is that routers are generally not as fast as
switches both in throughput and latency. Newer, expensive routers may
be, but your basic $200 model isn't going to forward traffic at the
speed of a switch. Plus, routers generally only have one port on each
subnet, so that is a bottleneck in itself. Layer-3 switches generally
are as fast as layer-2 switches but much more expensive.

....kurt

On 15 Aug 2006, in the Usenet newsgroup comp.os.linux.networking, in article
<(E-Mail Removed) om>, taihc wrote:

>I'm new to networking and have some questions on my mind that I hope
>can be answered.

http://www.ietf.org/rfc/rfc0950.txt

0950 Internet Standard Subnetting Procedure. J.C. Mogul, J. Postel.
August 1985. (Format: TXT=37985 bytes) (Updates RFC0792) (Also
STD0005) (Status: STANDARD)

http://tldp.org/guides.html

* The Linux Network Administrator's Guide, Second Edition
version: 1.1
authors: Olaf Kirch and Terry Dawson
last update: March 2000
ISBN: 1-56592-400-2
available formats:
1. HTML (read online)
2. HTML (tarred and gzipped package, 690k)
3. PDF (1.5MB)

>I understand that subnets make managing networks easier, but what
>benefits are there of having hosts in the same subnet as opposed to
>different subnets? why are some servers placed in the same subnets and
>some placed in different subnets when they all need to communicate with
>one another?

That's a great big "That depends" type of question. The original reason
for subnets was to break a large network (back when we were using Class A
now called a /8, Class B, now called a /16 - in addition to a Class C now
called a /24) down into manageable/usable size. The original Ethernet
(10Base5) had a limitation such that there could be no more than 65025
hosts on the wire (255 physical networks of 255 hosts each). In _practical_
terms, that was ludicrous - as all were effectively sharing the same wire
and only ONE computer can "talk" at a time. A "Class A" network had
2^24-2 (16,777,214) usable addresses, while a "Class B" had 65534 usable.
How were you supposed to get that many computers onto an Ethernet? (The
other technology - Token Ring - was even more restricted.) The answer was
to subnet - to break those "large" blocks into smaller ones. Thus, we used
a /22 (255.255.252.0 or 0xfffffc00) mask, which allowed 1022 hosts on each
subnet. (In fact, we never had more than about 750 on each.)

So, how to break things into manageable chunks. The answer is - it's up
to you. You might break things down based on one subnet per building, or
per floor, or per wing (physical). You might break it down according to
departments - sales, accounting, engineering (organizational) or just
about any other criteria. Hey, it's your network, do _your_ thing.

Now, some practical points. Traffic through a router is slower - it's a
"choke point" and remember, only one conversation at a time. Thus, you
want to put the file servers locally. Security - the original Ethernets
were a "party line", and anyone on the network could "hear" all conversations
on the network. Thus, you isolated "sensitive" systems to their own net. This
also carries over to the idea of having a DMZ or isolated network that will
run riskier (public accessible) applications/services. Traffic density -
you want to isolate "talkative" networks, making them smaller perhaps, so
that other systems can have a chance to "talk".

In about 1994, a new network appliance called a 'switch' became available.
This allowed you to further break down your subnets, because the switch only
routed traffic between source and destination ports. The Kalpana Etherswitch
had 16 ports. We broke our 10Base5 subnets into smaller sections, such that
there was no more than 50 hosts on any port, and the routers and file servers
each got their own port. Vast improvement in throughput, with no changes on
the individual hosts - for a mere US$7,000 each. (A 24 port dual speed
[10/100BaseT] switch today costs well under US$1000.)

Nearly all networks today use a switch technology. The problem is that these
are limited in the number of ports (288 port switches are available) and
bandwidth. Briefly, the switch has to have the horsepower to "repeat" some
significant fraction of the bits trying to go through it. Let's say you
have a ten port 100BaseT switch - the switch has to be able to listen to
the incoming port long enough to see where the packet is going (first six
bytes of the packet), set up an internal route from input port to output
port, and then _repeat_ the packet to this output port. Oh, and while you
are goofing off, there's another packet coming in on that other port that
needs to go "there" - pull yer finger out!!! Total bandwidth is more
expensive than ports - but neither one is cheap as dirt.

So - look at your network. What traffic is going where? What (if any)
security problems need to be (can be) taken care of. Perhaps you have only
a limited number of 'public' IP addresses, and need to look at those systems
that need to be publicly reachable, verses those that can use an RFC1918
address range and be NATed when they need to reach out. In short, what does
your network need.

Old guy

In article <(E-Mail Removed) om>,
"taihc" <(E-Mail Removed)> writes:
>
> I'm new to networking and have some questions on my mind that I hope
> can be answered.
>
> I understand that subnets make managing networks easier, but what
> benefits are there of having hosts in the same subnet as opposed to
> different subnets? why are some servers placed in the same subnets and
> some placed in different subnets when they all need to communicate with
> one another?

You've already got a couple of answers, so I won't repeat what's in them.
Instead, I'll add this: Setting up subnets enables you to put in layered
security. For instance, you can have firewalls between your subnets to
restrict what types of protocols can be used between them, or between any
of your subnets and the Internet at large. This can be particularly
important if you've got, say, Internet servers (mail servers, Web servers,
etc.) and systems that should not be accessible from the Internet at
large (local workstations, print servers, etc.). Put each type of system
on its own subnet and configure your router(s) with appropriate firewall
rules and you've greatly improved and/or simplified your network security
compared to leaving everything on one subnet and setting up rules for
individual systems.

Another reason to use multiple subnets is to support multiple types of
networking hardware. You might want to use different subnets for wired
computers vs. wireless access, for instance.

--
Rod Smith, (E-Mail Removed)
http://www.rodsbooks.com
Author of books on Linux, FreeBSD, and networking