John, some comments on your ATM notes.
While debates on bandwidth in the core may be moot (with a bandwidth glut coming,
who cares, I suppose), I note that the biggest portion of wasted (50%) bandwidth is
for self healing duplex sonet rings. I think the network will migrate to a self healing
mesh network, which will increase efficiency and reliability, but at some cost.
Both ATM and IP can work with that topology, though both requires a network of
ATM or IP switches, and ATM switches have historically been cheaper than IP.
But on to your comments.
Guys,
ATM as a desktop technology is dead. period.
Today's conventional wisdom would agree with you.
I don't really have anything that challenges that, except xDSL/ATM.
ATM as a core technology has no merit at speeds above OC3.
ATM does work as a technology above OC-3; by merit I assume you mean it isn't
suitable or cost effective. But I understood that the most common core deployment
was IP over ATM over SONET, at speeds up to OC-48. After OC-48 people are
talking about optical switching. This is only
my opinion, but as Jim Kayne notes, ATM sales are increasing. I would like add that
sales of ATM/SONET chips in companies like PMCS and TXCC and others, are very
strong (see their press release and quarterly statements), often doubling over the last year- which means that six months later, equipment will
be sold to service providers.
So if it doesn't work to the desk it must be for some other reason, and apparently no
one has told the service providers. My guess is inside the office people prefer IP
management, and don't like the over manage of ATM, and most 100 base T's systems
have bandwidth to spare.
The cell tax easily exceeds 15%. There is a reason that Cisco and Juniper rule the
core of ISP's - they do packet over sonet. The competition only did packet over atm
and missed the boat.
I presume this refers to the 5 byte overhead on 48 bytes payload? I noted in an earlier
post that small packets are not efficient in IP, and for something like voice over ip,"
"An VOIP packet contains a IPv6 header (60 bytes- v4 is 40 bytes), a UDP
header (8 bytes) , a RTP header (12 bytes) and a H.729a payload of 10 or so
bytes, and MPLS (4 bytes). This is 84 bytes for a 10 byte payload. Now even
H.729a is not quite as good quality as 'toll' quality voice, and it is about, I
remember a 10:1 compression algorithm We are almost at parity with TDM! in
bandwidth efficiency and getting lower quality."
ATM's 5 bytes to 48 payload sounds real good to me, even if you are only transmitting
32 bytes of uncompressed TDM payload, particularly if it maintains TDM quality.
Of course there are other layers of overhead, but IP seems to be adding those layers
as well.
As to your market analysis, I strongly agree with you that in a IP and ATM debate, the
best solution is to pick both. This is why Alcatel bought Newbridge, and Lucent
bought Ascend. If would also think FR and DWDM are mandatory. The problem
will be with vendors who only supply one and not all; they are buyout candidates, I
think.
Where does it fit - service provider carrier networks providing low speed (DS0 to
DS1) connectivity to large enterprises with branch operations. As the underlying
technology for DSL access and in similar environments.
Well, yes: the enterprise market. This is rather large. Most of it today is leased line,
followed by FR, followed by ATM. It is possible fiber based IP 10G can make an
inroads to this market, but most enterprises need voice as well as data, so you are
talking either VOIP, or two networks. The only think that will drive this, is massive use
of vide conferencing, in my opinion, and I don't know which is better for that: IP is my
guess.
You make a good point about DSL access- but I understand some xDSL uses IP
instead of ATM; however this last mile is still a good market for ATM, as you point
out.
If there is a fiber glut in the core, then the next area of growth will be the
metro/enterprise, and the last mile. So there may be no reason to upgrade the installed
ATM of the core if DWDM gives you more bandwidth. So it will stay
IP/ATM/SONET perhaps. At least that is what some on this thread of noted, for the
near term.
As to the vaunted QOS properties of ATM - most commercial services do not
provide them. You are lucky to get congestion notification on many provider
networks.
In the lab and on paper ATM shines. In the real world its relegated to a niche.
My understanding is that ATM is beloved of the operations staff of service providers.
Is your point that congestion notification doesn't reach the end user? I didn't know
that.
Yes this niche is growing, but don't read too much into it.
John, some comments on your ATM notes.
While debates on bandwidth in the core may be moot (with a bandwidth glut coming,
who cares, I suppose), I note that the biggest portion of wasted (50%) bandwidth is
for self healing duplex sonet rings. I think the network will migrate to a self healing
mesh network, which will increase efficiency and reliability, but at some cost.
Both ATM and IP can work with that topology, though both requires a network of
ATM or IP switches, and ATM switches have historically been cheaper than IP.
But on to your comments.
Guys,
ATM as a desktop technology is dead. period.
Today's conventional wisdom would agree with you.
I don't really have anything challenge that, except xDSL/ATM.
ATM as a core technology has no merit at speeds above OC3.
ATM does work as a technology above OC-3; by merit I assume you mean it isn't
suitable or cost effective. But I understood that the most common core deployment
was IP over ATM over SONET, at speeds up to OC-48. After OC-48 people are
talking about optical switching using DWDM, which can carry anything. This is only
my opinion, but as Jim Kayne notes, ATM sales are increasing. I like to point out the
sales of ATM/SONET chips companies, like PMCS and TXCC and others, are very
strong, doubling over the last year- which means that six months later, equipment will
be sold to service providers.
So if it doesn't work to the desk it must be for some other reason, and apparently no
one has told the service providers. My guess is inside the office people prefer IP
management, and don't like the over manage of ATM, and most 100 base T's systems
have bandwidth to spare.
The cell tax easily exceeds 15%. There is a reason that Cisco and Juniper rule the
core of ISP's - they do packet over sonet. The competition only did packet over atm
and missed the boat.
I presume this refers to the 5 byte overhead on 48 bytes payload? I noted in an earlier
post that small packets are not efficient in IP, and for something like voice over ip,"
"An VOIP packet contains a IPv6 header (60 bytes- v4 is 40 bytes), a UDP
header (8 bytes) , a RTP header (12 bytes) and a H.729a payload of 10 or so
bytes, and MPLS (4 bytes). This is 84 bytes for a 10 byte payload. Now even
H.729a is not quite as good quality as 'toll' quality voice, and it is about, I
remember a 10:1 compression algorithm We are almost at parity with TDM! in
bandwidth efficiency and getting lower quality."
ATM's 5 bytes to 48 payload sounds real good to me, even if you are only transmitting
32 bytes of uncompressed TDM payload, particularly if it maintains TDM quality.
Of course there are other layers of overhead, but IP seems to be adding those layers
as well.
As to your market analysis, I strongly agree with you that in a IP and ATM debate, the
best solution is to pick both. This is why Alcatel bought Newbridge, and Lucent
bought Ascend. If would also think FR and DWDM are mandatory. The problem
will be with vendors who only supply one and not all; they are buyout candidates, I
think.
Where does it fit - service provider carrier networks providing low speed (DS0 to
DS1) connectivity to large enterprises with branch operations. As the underlying
technology for DSL access and in similar environments.
Well, yes: the enterprise market. This is rather large. Most of it today is leased line,
followed by FR, followed by ATM. It is possible fiber based IP 10G can make an
inroads to this market, but most enterprises need voice as well as data, so you are
talking either VOIP, or two networks. The only think that will drive this, is massive use
of vide conferencing, in my opinion, and I don't know which is better for that: IP is my
guess.
You make a good point about DSL access- but I understand some xDSL uses IP
instead of ATM; however this last mile is still a good market for ATM, as you point
out.
If there is a fiber glut in the core, then the next area of growth will be the
metro/enterprise, and the last mile. So there may be no reason to upgrade the installed
ATM of the core if DWDM gives you more bandwidth. So it will stay
IP/ATM/SONET perhaps. At least that is what some on this thread of noted, for the
near term.
As to the vaunted QOS properties of ATM - most commercial services do not
provide them. You are lucky to get congestion notification on many provider
networks.
In the lab and on paper ATM shines. In the real world its relegated to a niche.
My understanding is that ATM is beloved of the operations staff of service providers.
Is your point that congestion notification doesn't reach the end user? I didn't know
that.
Yes this niche is growing, but don't read too much into it.
Pretty neat trick, for a dead technology!
Bottom line: Who wins on IP vs. ATM?
Actually, the winner is DWDM. Who would have thought circuit switching would triumph
over packets or cells! |
