Link@GUNTER-ADAM.ARPA (04/05/88)
I'm getting the legwork started to connect 42 local area (802.3) networks up to the MILNET and have run into a hitch. A single Class B network would have ample address space for the hosts on all 42; however, each of my networks is geographically separated. I could request Class C networks, but several of my nets will have more than the allowed 254 (256) hosts. Sue at the NIC (HOSTMASTER) sent me INTERNET-NUMBER-TEMPLATE.TXT, which is the application for net numbers and referenced me to RFC 950, the Subnetting procedures. Neither of these really answer my questions; the template strongly recommends subnetting if you have multiple lans and more than a 100 hosts and RFC 950 discusses that subnetting. All good information, in fact we will be running local subnets at most locations (all those that have more than a single baseband segment) and will use a local bit mask on the bridges. My problem is we'll have 42 separate connections (gateways) to the MILNET and it's my understanding that I can't break up a Class B (or any class net, for that matter) network into subnets across the DDN. Because of the reasoning shown above, I'm planning on requesting a block of 42 Class B network numbers. I realize that Class C would suffice for many of my nets, but allowing for growth, most of them should exceed the address space of a Class C net in a fairly short time. Of course if I've made some glaring error in reasoning (like not knowing of a graceful way to cram more than 256 hosts into an 8-bit address space) feel free to flame me. Otherwise, I'd appreciate any pointers anybody out there can give me; I like pats on the back too... I'm also looking for the gateways to connect all these lans to the MILNET. I know a few vendors, but know someone would be upset if they didn't get a copy of our IFB and they had a product that satisfy the requirement. Please address all replies directly to me; I might miss them in filtering through the rest of the TCP-IP mail. I'll summarize for the world. Thanks. |====================================================================| | Link Verstegen Network Solutions, Inc. (NSI) | | Lead Hardware Engineer 4350 Will Rogers Parkway, Suite 100 | | Oklahoma City, OK 73064 | | Link@Gunter-Adam.ARPA (405)942-8884 | |====================================================================| -------
STJOHNS@SRI-NIC.ARPA (04/06/88)
You are doing exactly the right thing... 42 class B addresses. BUT... if you decide to install trunkc between your various gateways, you could treat all the networks served by those gateways as part of the same class B subnetted network. Mike St Johns, DDN Program
brescia@PARK-STREET.BBN.COM (Mike Brescia) (04/06/88)
<from STJOHNS>
You are doing exactly the right thing... 42 class B addresses.
Since the idea of subnetting is to hide the distinction among the subnets from
the rest of the routers, you can use a single net which you subnet. The
penalty you pay is to have the other routers send to an arbitrary one of your
42 gateways in order to reach a particular subnet. This means you must be
prepared to take a packet addressed to subnet #27 in on the gateway for subnet
#33 and forward it to the proper your other gateway for subnet #33. You may
have better (faster) connections inside your domain than on the DDN, so it may
not be too painful in user delay or DDN overload.
If the source of the packet is a host on the DDN, rather than a gateway, then
your gateway can send an ICMP redirect-host to get further packets to the
proper gateway [flames about hosts which do not listen to redirect-host go to
those hosts].
This example has come up often in the past, but usually the number of sites
has been around 2 rather than 42. For the cases where the sites have no
separate connection, they used separate nets. For the cases where the
sites had backdoor trunks, or a bridged ethernet, they used a single net, and
paid the extra penalty of sending traffic over their internal trunk that came
in on the 'wrong' gateway.
Mike