Transcript Q and A slides

Q and A, Ch. 21
CS332, Spring 2016
Victor Norman
Dotted-decimal notation
Q: I still don't understand dotted decimal. Are
we supposed to know how to convert from 32
bit to dotted decimal? If so how do I even begin
to do that?
A: An IP address is 4 bytes = 32 bits. One byte is
a decimal number from 0 to 255. So, you
convert each byte of the address to decimal and
put a . between each part.
IP Address Parts
• Every IP address has two (or 3) parts:
– Network part: uniquely identifies the network
universally.
– Host part: uniquely identifies the host on the network.
• Why is this done?
– Routing is done on the network part of an IP address.
– For scalability.
• Part of the host part can be used as a subnet
part, within an organization.
Classful IP Addressing
•
•
•
•
(the old way)
Figure 21.1
4 kinds of addresses
Didn’t have to specify the mask, because you
could figure it out from the address.
• Exercise: what class does 200.201.202.203
belong to?
Address masks
Q: Can you explain address masks a bit more?
A: An address mask indicates with 1 bits the part of
an address that is the network part. The 1 bits are
all consecutive and at the “left-most” part of the
address.
• Can be shown as /n  first n bits are 1s  which
means the first n bits are the network part.
• E.g., Class C address  24-bit network part 
255.255.255.0  /24
NOTE: masks only needed in classful addressing to
indicate the subnet part.
Subnet Addressing
• With classful addressing, if you got a class B address, you
had 1 LAN with 2^16 hosts on it.
• What if you wanted 2 LANs? (Why would you?)
• Internally in your network, you could subdivide your LAN
into multiple LANs (or subnets) using subnetting.
• “Borrow” some bits of the host part to indicate the subnet.
• 1 bit  2 LANs; 2 bits  4 LANs; etc.
• Now, you have to route between LANs in your organization.
• Now, every machine in your network has to know the
netmask for the network  so that it can figure out if
another IP address is on its LAN or on another LAN.
How to tell if a machine is on a
network?
• A machine with address addr is on network N
with mask M if addr & M == N
– N is 32-bit network address (host part all 0s)
– & is binary AND
– Could read this: if the network part of addr equals the
network N.
• Exercise: is 153.106.129.33 on network
153.106.128 / 24 ?
– How about 153.106.128 / 17?
• Q: does every interface on a LAN have to have the
same mask?
Exercise
• BeatHope.com, LLC, needs IP addresses for 500
computers (and we still live in a classful
addressing world)
– What class of IP addresses should they request?
• They want 3 subnets, for privacy/security:
–
–
–
–
30 machines on one subnet,
50 machines on another,
420 machines on the last.
What subnetting scheme should they use?
• How does “the world” see BeatHope.com’s
network? I.e., what mask does “the world” use?
Classless Addressing
• What is the problem with having only class A, B, or C
address blocks?
– If you need, say, 256 hosts (which is > 254), you need class B,
which means ~65,200 wasted IP addresses…
• What is the solution?
– Assign blocks of IP addresses to organizations based on any
number of bits in the mask. (instead of just 8, 16, or 24 bits)
– e.g., 28 bit mask  2^4 – 2 hosts = 14 hosts.
– e.g., 23 bit mask  2 ^ 9 – 2 hosts = 510 hosts.
• Means many 23-bit IP address blocks can be given out to
many companies, instead of 1 16-bit address block.
• Means far fewer wasted IP addresses and far more # of
networks.
Example
• Tiny Corp wants a network with max 14 hosts.
• Network provider gives them 154.100.1.0/28
– first 3 bytes have to be 154.100.1, top half of 4th byte has
to be all 0s. Hosts in the network will be from 154.100.1.1
– 154.100.1.14. Bcast address is 154.100.1.15 (all host bits
are 1s).
• Joe’s Pizza asks for and gets network 154.100.1.16/28.
– top half of 4th byte is 0001. Hosts are 154.100.1.17 –
154.100.1.30.
• BubbleGumRUs has network 154.100.1.32/28…
• 16 networks with 154.100.1/28 can be allocated
Do classes matter anymore?
Q: In classless routing, do classes matter any
more?
A: Sort of. If you have what used to be a class B
address (16-bit network part), then your
network must have a mask of at least 16 bits.
You can’t just have 8 bits, because there are 2^8
networks with the same top 8 bits.
In other words, from looking at an address
alone, you can tell the mask must be >= some #.
Transition to CIDR
Q: How did the transition to CIDR happen? What
had to be done?
A: Subnet masks were used internally within an
organization already. With CIDR, masks needed to
be used in ISPs (i.e., the Internet backbone) for
proper routing.
So, network masks needed to be
advertised/distributed to all ISP routers along with
the network numbers themselves.
So, change in route propagation protocols (OSPF,
RIP, BGPv4, etc.)
Watch out!
• One problem with masks that aren’t 8, 16, or 24
bits is that IP addresses are still written
8-bits . 8-bits . 8-bits . 8-bits.
• E.g., In CS at Calvin, we have machines on
153.106.116.* and 153.106.117.* that are on the
same LAN… Hard to tell that from looking at the
addresses alone.
– 116 = 0111 0100, 117 = 0111 0101
– mask is /23 – 255.255.254.0.
– last 9 bits of the IP address are the host part.
Reuse of addresses
Q: Can IP addresses be reused within a network?
A: Yes. IP addresses within an organization are
usually assigned with DHCP. These addresses
are “leased”, and can be taken back and given to
another machine.
Address depletion, sales, etc.
Q: Can IP addresses be sold?
A: I thought it was “no”, but it turns out it is
“yes”.
http://www.gtri.com/how-to-buy-or-sell-ipv4addresses/
https://www.arin.net/resources/index.html
https://www.iana.org/numbers
Special addresses
• Network address:
– all host bits in the address are 0s.
– Used in routing tables, etc.
• Directed broadcast address:
– all host bits in the address are 1s.
– when used as a destination address in a packet, it
means the packet is routed to the destination
network and then broadcast on it.
– Not done by routers today.
Special addresses (2)
• Limited broadcast address
– 32 1 bits.
– As a destination address, means send to all
machines on the local Layer 2 network.
• Loopback address:
– 127.0.0.0/8 – typically 127.0.0.1
– Used to test your local TCP/IP stack or to send
some packet back to yourself so another task can
process it.
“Joke shirt”
Multiple IP addresses on one
machine?
Q: Is it possible for a host to have more than one IP
address?
A: Yes! If a host has multiple routing interfaces then
it must have multiple IP addresses.
• A router/host has to decide how to get a packet
to its destination: can the packet be sent directly
to its destination or does it have to go through a
router to be forwarded?
• So, each interface has to have an IP address (and
know the netmask) for the network it is on.
IP broadcast
Q: How much is IP broadcast used?
A: Limited broadcast is used when you have to
reach every/any machine on ”this” network. It
is used for …
• DHCP request
• ARP request
Old slides
Universal Addressing Scheme
Q: Apart from the fact that arbitrary pairs of
application programs can communicate without
knowing the type of network hardware, what
are the other needs for a universal addressing
scheme?
A: That’s about the only thing I can think of…
Dotted-decimal notation
Q: I still don't understand dotted decimal. Are
we supposed to know how to convert from 32
bit to dotted decimal? If so how do I even begin
to do that?
A: An IP address is 4 bytes = 32 bits. One byte is
a decimal number from 0 to 255. So, you
convert each byte of the address to decimal and
put a . between each part.
Non-computers have IP addresses?
Q: The chapter talks about computers having IP
addresses but do other devices like printers have
them as well?
A: Yes! Anything that needs to talk on the
network needs to have an IP address.
No hierarchical structure?
Q: Could the Internet have been designed with
no hierarchical structure of IP addresses so none
would be wasted?
A: It could but it wouldn’t scale. Packets are
routed based on their network part only. This
allows one routing entry for hundreds or
thousands of hosts. Without a hierarchy, you’d
have to have an entry for every host.
Suffix not needed?
Q: If you apply a mask to an IP address, you only
see the prefix right? Why? Don't you need to
know the suffix as well?
A: The mask is applied to the address in a host
or router when deciding how to route the
packet. It is not applied anywhere else. And,
routing is based on the network an address
belongs to, so to make this decision, the
host/router doesn’t need to know the suffix.
Need for classless addressing
Q: Can you explain the need for classless
addressing?
A: IANA began to see that lots of addresses were
being wasted. If your company needed 300
addresses, you couldn’t get a class C address
(254 addresses max), so you got a class B, with
65534 addresses, most of which went unused.
And, the number of class B networks was going
down fast…
Lots of IP addresses available?
Q: Aren't there so many IP addresses available,
that it would never matter if we were wasteful?
A: No! The last set of addresses was given out
by the IANA to a regional address registrar last
year. We are now out of IP addresses.
More memory for CIDR?
Q: Did it use a lot more memory to start storing a
32 bit mask along with every 32 IP address?
A: The question is, where is this extra memory
required? The IP packet header didn’t change at all.
Routing tables didn’t change either – because
subnetting was available before CIDR. The only
thing that changed is routing protocols, and they
take more memory now, and the fact that ISPs can
hand out non-/8, /16, and /24 blocks of addresses.
IP Address Reuse
Q: Are IP addresses reused, and does a server
store what IP addresses are being used and then
know when one is no longer being used?
A: IP addresses are re-used in a network that
assigns them via DHCP. And, they could be
reused in the Internet when one company’s
address block is not used anymore.
Switching algorithm
• in memory have a table that maps between port and
mac address.
• receive frame on port n, with source mac smac and
dest mac dmac.
• look up smac in table. Add entry if not there. Update
entry if there, and port is not n.
• if dmac is bcast (all 1s), send to all ports, except n.
• look up dmac in table: if not there, send frame to all
ports except n. If there, send frame to only port
associated with dmac.
• go through all entries in table and remove old ones.
Practice questions
• Need a network with 2000 hosts. What class
network would you get in the old days when
we did classful networking? With classless
addressing, your ISP would give you a network
with what mask?
Practice questions (2)
• For network 200.201.220.16/28:
– how many hosts can you have?
– what is the broadcast address?
– what is highest IP address?
Practice (3)
• You get 153.106.96.0/20 from your ISP.
– How many bits for host part?
– What is the limited broadcast address?
• You want 16 subnets, so what is your internal
subnet mask?
• What is the lowest host IP address for subnet
3 (starting with 0)?