Question rephrase
From: Iceman (iceman_to_the_max_at_yahoo.com)
Date: 10/26/04
- Next message: dougga: "Re: NFS through a wireless router"
- Previous message: Boris Glawe: "Re: root over nfs - performance problems"
- In reply to: Joachim Maeland: "Re: Any other way for non-static IPs?"
- Next in thread: Moe Trin: "<LONG>Re: Question rephrase"
- Reply: Moe Trin: "<LONG>Re: Question rephrase"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: 26 Oct 2004 14:39:10 -0700
> Anyway: What are you trying to accomplish and why do you insist on
> going in this direction?
I apologize for making such a big deal about a small thing, but it's
more a style and clarity question.
* Problem setup
You have 100 machines (Linux, Windows, MacOS, whatever...). They all
have ONE ethernet adapter. Let's concetrate on 5 Linux machines, which
have their respective eth0 adapters. I will say "some server" instead
of DHCP server here, because this might be a question where answer
could be Dynamic DNS or any other type of name service. Assume that
all 5 Linux machines have that server (let's call it NS - name server)
up and running, one is a master, four are slaves and that any of them
can become master if current master fails (thunderbolt, whatever...).
When any host (even one of the 5 Linuxes running the server) turns on,
it tries to get it's IP. The only unique is their ethernet MAC
address. So it sends MAC to FF:FF:FF:FF:FF:FF broadcast. There is
currently only one master
NS server, which responts. This is something like the DHCP's
DISCOVER/OFFER/REQUEST/ACK protocol. OK, so now it has it's address
and it can work OK.
However, if there was no NS server available, why not configure
itself? Since you are yourself a NS, you are able to do it and, in the
same time, become a master NS for that network.
A note: this is the simmilar to TokenRing networks, where there is a
protocol for defining the master computer for the network. Of course,
this is just to define master for the ethernet (ISO OSI 2) layer, not
IP (ISO OSI 3).
* Pros
- No static addresses for NS server configurations. Administrators
wouldn't need to configure anything, but a network IP, netmask and
range of IP addresses. They could say: you are a possible NS (there
may be other NS servers) for the 192.168.0.0 network with
255.255.255.0 netmask, you have a range 192.168.0.10 - 192.168.0.100
to lease.
- If eth0 is given such network IP, netmask and range, NS which is
joining on a network without a master could say: OK, no master, so
noone can give me the IP. Since I am a potential NS, I will become a
master on that NETWORK (so I still don't have an IP ADDRESS for my
eth0) and I will issue a client request ON THAT NETWORK again. Since
now there is a master now (itself, but why not?), it will issue an IP
for the client (itself again) and it will get a valid IP.
- No static NS addresses for clients. Clients would only join a
network and look for a NS server. No static DNS, WINS IPs or whatever
- just broadcast and you will get it from the current master NS. Many
NS possible, at any available IP on that network as they wish.
* Cons
- Some security issues may be at stake here, but that's not what I am
"into" right now. This is a potentially big thing in many network, so
a "contra" point here.
- May be slow.
* The real problem
Back to my apology, I have only 2 machines (OK, it's a big story for
such a "huge" network, sorry), one has WinXP/Linux (A) on it and the
other WinXP (B). Static IP works just fine. I wanted to make it
dynamic. If I start Linux (A), I must make it static IP in order for
it to be a DHCP server, right? So what is so dynamic about this? If I
join the other WinXP (B), it will get a dynamic IP. 50% is not
enogugh.
But a better example: assume I had two Linux machines, both of which
could be DHCP servers. How to do it? Since I need a DHCP server to be
on a static IP, both would be on static addresses. 0% is nothing.
* Questions (hopefully clear now)
Is there a way to name 4 machines Dragon, Torpedo, Mushroom and
Turtle, give their network IP, netmask and range of addresses on that
network they can occupy and the system does the rest? I don't want to
care about their IPs, but I want to control them enough. If an upper
authority gave me a network range, I must obey to that, but nothing
else. I have network from "here" to "there", it's mine and I want a
way to fit my 4 computers (and maybe afterwards add up to 60 laptops,
network printers, etc.) without caring who, when, why, ... assigns
IPs. I want simple server configuration. I want to be able to simply
rip of the above 4 machines (all NS severs) and add 15 NS servers
more. I don't want to configure them again by assigning 15 static IPs
- that is boring, error-prone and certainly not DYNAMIC.
* More questions
- Is my thinking bad or good? I.e. do you agree in common?
- Is there anything working in practice like this? Is Dynamic DNS what
I am looking for? Can DHCP do the above?
- If not, any ideas why not - what are the problems for it not being
like this?
- How do you manage (if you do) large networks? Static DNS is also
very demanding, as I see.
* That's all, folks!
Thanks again for your time and answers!
Iceman
P.S.: If you read all this, then you:
a) have a lot of time to waste,
b) are very interested for the topic,
c) are a very good person,
d) (... I surely missed something here, no? ...)
- Next message: dougga: "Re: NFS through a wireless router"
- Previous message: Boris Glawe: "Re: root over nfs - performance problems"
- In reply to: Joachim Maeland: "Re: Any other way for non-static IPs?"
- Next in thread: Moe Trin: "<LONG>Re: Question rephrase"
- Reply: Moe Trin: "<LONG>Re: Question rephrase"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
