s*****g
发帖数: 1055 | |
s*****g
发帖数: 1055 | 2 As the leading network equipment vendor, Cisco management should probably
ask itself and Google why Google did not buy (a lot) Cisco equipment to
build one of the largest network in the world. If Cisco's current loyal
customers start to follow Google, Cisco (and other traditional network
vendors) will be in big trouble.
【在 s*****g 的大作中提到】
: http://cseweb.ucsd.edu/~vahdat/papers/b4-sigcomm13.pdf
: http://www.networking-forum.com/viewtopic.php?f=46&t=29803
|
v**n
发帖数: 951 | 3 我觉得你多虑了。 google的case不是一般的特殊。
google是第一个也是最大的openflow的production network。但是这个特例会想烽火一
样燎原吗？我觉得不会。除非google以后把所有的SP全部吃掉或者干掉。当然这个可能
会发生。。。
【在 s*****g 的大作中提到】
: As the leading network equipment vendor, Cisco management should probably
: ask itself and Google why Google did not buy (a lot) Cisco equipment to
: build one of the largest network in the world. If Cisco's current loyal
: customers start to follow Google, Cisco (and other traditional network
: vendors) will be in big trouble.
|
z**r
发帖数: 17771 | 4 以后网络hybrid的可能性更大，Cisco ONE看着要比OF更全面，就跟Cisco其他产品和
Google自己的switch比一样，适合Google的不见得适合其他人
【在 v**n 的大作中提到】
: 我觉得你多虑了。 google的case不是一般的特殊。
: google是第一个也是最大的openflow的production network。但是这个特例会想烽火一
: 样燎原吗？我觉得不会。除非google以后把所有的SP全部吃掉或者干掉。当然这个可能
: 会发生。。。
|
f*****m
发帖数: 416 | 5 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
) |
z**r
发帖数: 17771 | 6 高端黑呀。
【在 f*****m 的大作中提到】
: 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
: )
|
L******t
发帖数: 1985 | 7 What did those customers want from OpenFlow?
Except Google, who is one of a kind, what most other consumers of OpenFlow
want is flow monitoring/redirecting functionality, which is a good use case
of OpenFlow. Many a vendors are busy with implementing such featurettes on
various platforms per customers' demands.
【在 f*****m 的大作中提到】
: 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
: )
|
wu
发帖数: 28 | 8 一直没搞懂OF能搞定什么？我指的的是针对企业用户。到底有什么样的企业会对数据层
有编程需要？ |
x*********n
发帖数: 28013 | 9 我看是没有意义，
OF的意义就是省cost，这个省得是特定情况才行，估计也省不了多少。
但是IT嘛，过阵子有新东西喊一喊总是有好处的。
【在 wu 的大作中提到】
: 一直没搞懂OF能搞定什么？我指的的是针对企业用户。到底有什么样的企业会对数据层
: 有编程需要？
|
x*********n
发帖数: 28013 | 10 我看是没啥好担心的，google这种公司，从management看就是这样，
没事出一个idea，然后弄一弄，迷糊一下投资者，
我看到是google fibre现在投入很大，未来5-10年前景看好。 |
|
|
s*****g
发帖数: 1055 | 11 SDN/OF in multi-tenant Data Center environment promises rapid/scalable
network deployment. I don't see how much SDN/OF can help in campus/LAN
environment, I don't think cost reduction is a big selling point of SDN/OF,
it is more about the problems SDN/OF can solve.
What I am really interested in SDN/OF is its potential in WAN networking,
traditional networking separates control plan and forwarding plan only at
device level, control plan does not even how forwarding plan is doing (i.e.
link utilization) on a particular device, let along end to end how
forwarding plan is behaving, this makes the applications across WAN/Internet
suffer. Current solution for the above problem is to manul traffic-
engineering (PBR, BGP communities etc) as a kluge, but this is too
cumbersome and hard to maintain.
Having a central traffic control system (if you own transport) solves the
above problem naturally. I yet to think through how to make Google's
implementation also work in overlayed (say, GRE/Internet) network - if
possible at all.
【在 x*********n 的大作中提到】
: 我看是没有意义，
: OF的意义就是省cost，这个省得是特定情况才行，估计也省不了多少。
: 但是IT嘛，过阵子有新东西喊一喊总是有好处的。
|
f*****m
发帖数: 416 | |
m**k
发帖数: 290 | 13 nice article, thanks for sharing!
SDN目前只有google这样的公司玩得起吧。
b4 貌似不是纯种的SDN。它还是需要在data plane跑BISIS。TE是附加在其上的一层。
感觉就是(或者还不如 MPLS with centralized TE.
另外b4的前景如何有待观察。如果google只是自己用，升级换代研发测试开销巨大，
跟其它解决方案比起来不一定有优
,
.
Internet
【在 s*****g 的大作中提到】
: SDN/OF in multi-tenant Data Center environment promises rapid/scalable
: network deployment. I don't see how much SDN/OF can help in campus/LAN
: environment, I don't think cost reduction is a big selling point of SDN/OF,
: it is more about the problems SDN/OF can solve.
: What I am really interested in SDN/OF is its potential in WAN networking,
: traditional networking separates control plan and forwarding plan only at
: device level, control plan does not even how forwarding plan is doing (i.e.
: link utilization) on a particular device, let along end to end how
: forwarding plan is behaving, this makes the applications across WAN/Internet
: suffer. Current solution for the above problem is to manul traffic-
|
d****i
发帖数: 1038 | 14 谢楼主发的第一个连接。 只有一个感想，总算见到活的了。
文章非常具体的讲了SDN的实现，回答了我以前一直不太确定的事情，很有启发性
也可以看出在未来几年内sdn对传统网络和网络设备的冲击是有限的。
cisco 的one如果作好也可以做不错的过渡。
【在 s*****g 的大作中提到】
: http://cseweb.ucsd.edu/~vahdat/papers/b4-sigcomm13.pdf
: http://www.networking-forum.com/viewtopic.php?f=46&t=29803
|
d****i
发帖数: 1038 | 15 Re this.
I think the real selling point of OF and SDN is the manageability it bring
to table. I don't mean the traditional network management here, which
actually is too focused on device management.
The 大棋 SDN and OF brings in is the focus of network management on the real
network level/end to end level and even at per flow level. And the
management includes all aspect of the network from data plane to control
plane to management plane.
but from this article it actually shows even google doesn't have a general
solution for such a 大棋. Its SDN can only solve or partly solve its own
problem.
,
.
Internet
【在 s*****g 的大作中提到】
: SDN/OF in multi-tenant Data Center environment promises rapid/scalable
: network deployment. I don't see how much SDN/OF can help in campus/LAN
: environment, I don't think cost reduction is a big selling point of SDN/OF,
: it is more about the problems SDN/OF can solve.
: What I am really interested in SDN/OF is its potential in WAN networking,
: traditional networking separates control plan and forwarding plan only at
: device level, control plan does not even how forwarding plan is doing (i.e.
: link utilization) on a particular device, let along end to end how
: forwarding plan is behaving, this makes the applications across WAN/Internet
: suffer. Current solution for the above problem is to manul traffic-
|
s*****g
发帖数: 1055 | 16 I think Cisco ONE's approach is different from what Google is doing, Cisco
ONE is more focus on providing APIs for network devices configuration, it
does not provide a central traffic control system that can take application
requirements and translate them to network device forwarding table change.
I also don't think this new idea of networking would benefit most medium to
small networks, but ISPs should take a close look at it, as they will be
able to increase their expensive long haul links' utilization and at the
same time provide VPN-like Diff services to customers.
I don't understand how the out-of-band network is built for Openflow across
WAN, I guess it still depends on in-band network built by traditional
networking, thought?
【在 d****i 的大作中提到】
: 谢楼主发的第一个连接。 只有一个感想，总算见到活的了。
: 文章非常具体的讲了SDN的实现，回答了我以前一直不太确定的事情，很有启发性
: 也可以看出在未来几年内sdn对传统网络和网络设备的冲击是有限的。
: cisco 的one如果作好也可以做不错的过渡。
|
z**r
发帖数: 17771 | 17 openflow is equivalent to just one part of the Cisco ONE, API is for
northbound which is callef ONEPK.
application
to
across
【在 s*****g 的大作中提到】
: I think Cisco ONE's approach is different from what Google is doing, Cisco
: ONE is more focus on providing APIs for network devices configuration, it
: does not provide a central traffic control system that can take application
: requirements and translate them to network device forwarding table change.
: I also don't think this new idea of networking would benefit most medium to
: small networks, but ISPs should take a close look at it, as they will be
: able to increase their expensive long haul links' utilization and at the
: same time provide VPN-like Diff services to customers.
: I don't understand how the out-of-band network is built for Openflow across
: WAN, I guess it still depends on in-band network built by traditional
|
d****i
发帖数: 1038 | 18 today most vendors are still focusing on implementing the agent part of SDN.
Cisco ONE is same, as a counterpart of OFA, to provide more flexible
programmable way for current IOS/NXOS based cisco devices.
I am not sure whether cisco plans or is doing something at controller level
though I think the big money of SDN actually will be on the controller side/
network OS and app side.
application
to
across
【在 s*****g 的大作中提到】
: I think Cisco ONE's approach is different from what Google is doing, Cisco
: ONE is more focus on providing APIs for network devices configuration, it
: does not provide a central traffic control system that can take application
: requirements and translate them to network device forwarding table change.
: I also don't think this new idea of networking would benefit most medium to
: small networks, but ISPs should take a close look at it, as they will be
: able to increase their expensive long haul links' utilization and at the
: same time provide VPN-like Diff services to customers.
: I don't understand how the out-of-band network is built for Openflow across
: WAN, I guess it still depends on in-band network built by traditional
|
s*****g
发帖数: 1055 | |
s*****g
发帖数: 1055 | 20 As the leading network equipment vendor, Cisco management should probably
ask itself and Google why Google did not buy (a lot) Cisco equipment to
build one of the largest network in the world. If Cisco's current loyal
customers start to follow Google, Cisco (and other traditional network
vendors) will be in big trouble.
【在 s*****g 的大作中提到】
: http://cseweb.ucsd.edu/~vahdat/papers/b4-sigcomm13.pdf
: http://www.networking-forum.com/viewtopic.php?f=46&t=29803
|
|
|
v**n
发帖数: 951 | 21 我觉得你多虑了。 google的case不是一般的特殊。
google是第一个也是最大的openflow的production network。但是这个特例会想烽火一
样燎原吗？我觉得不会。除非google以后把所有的SP全部吃掉或者干掉。当然这个可能
会发生。。。
【在 s*****g 的大作中提到】
: As the leading network equipment vendor, Cisco management should probably
: ask itself and Google why Google did not buy (a lot) Cisco equipment to
: build one of the largest network in the world. If Cisco's current loyal
: customers start to follow Google, Cisco (and other traditional network
: vendors) will be in big trouble.
|
z**r
发帖数: 17771 | 22 以后网络hybrid的可能性更大，Cisco ONE看着要比OF更全面，就跟Cisco其他产品和
Google自己的switch比一样，适合Google的不见得适合其他人
【在 v**n 的大作中提到】
: 我觉得你多虑了。 google的case不是一般的特殊。
: google是第一个也是最大的openflow的production network。但是这个特例会想烽火一
: 样燎原吗？我觉得不会。除非google以后把所有的SP全部吃掉或者干掉。当然这个可能
: 会发生。。。
|
f*****m
发帖数: 416 | 23 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
) |
z**r
发帖数: 17771 | 24 高端黑呀。
【在 f*****m 的大作中提到】
: 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
: )
|
L******t
发帖数: 1985 | 25 What did those customers want from OpenFlow?
Except Google, who is one of a kind, what most other consumers of OpenFlow
want is flow monitoring/redirecting functionality, which is a good use case
of OpenFlow. Many a vendors are busy with implementing such featurettes on
various platforms per customers' demands.
【在 f*****m 的大作中提到】
: 上次开会碰到一位，他的工作就是帮助客户跑OF. 这些用户跑了之后就死心了 :-
: )
|
wu
发帖数: 28 | 26 一直没搞懂OF能搞定什么？我指的的是针对企业用户。到底有什么样的企业会对数据层
有编程需要？ |
x*********n
发帖数: 28013 | 27 我看是没有意义，
OF的意义就是省cost，这个省得是特定情况才行，估计也省不了多少。
但是IT嘛，过阵子有新东西喊一喊总是有好处的。
【在 wu 的大作中提到】
: 一直没搞懂OF能搞定什么？我指的的是针对企业用户。到底有什么样的企业会对数据层
: 有编程需要？
|
x*********n
发帖数: 28013 | 28 我看是没啥好担心的，google这种公司，从management看就是这样，
没事出一个idea，然后弄一弄，迷糊一下投资者，
我看到是google fibre现在投入很大，未来5-10年前景看好。 |
s*****g
发帖数: 1055 | 29 SDN/OF in multi-tenant Data Center environment promises rapid/scalable
network deployment. I don't see how much SDN/OF can help in campus/LAN
environment, I don't think cost reduction is a big selling point of SDN/OF,
it is more about the problems SDN/OF can solve.
What I am really interested in SDN/OF is its potential in WAN networking,
traditional networking separates control plan and forwarding plan only at
device level, control plan does not even how forwarding plan is doing (i.e.
link utilization) on a particular device, let along end to end how
forwarding plan is behaving, this makes the applications across WAN/Internet
suffer. Current solution for the above problem is to manul traffic-
engineering (PBR, BGP communities etc) as a kluge, but this is too
cumbersome and hard to maintain.
Having a central traffic control system (if you own transport) solves the
above problem naturally. I yet to think through how to make Google's
implementation also work in overlayed (say, GRE/Internet) network - if
possible at all.
【在 x*********n 的大作中提到】
: 我看是没有意义，
: OF的意义就是省cost，这个省得是特定情况才行，估计也省不了多少。
: 但是IT嘛，过阵子有新东西喊一喊总是有好处的。
|
f*****m
发帖数: 416 | |
|
|
m**k
发帖数: 290 | 31 nice article, thanks for sharing!
SDN目前只有google这样的公司玩得起吧。
b4 貌似不是纯种的SDN。它还是需要在data plane跑BISIS。TE是附加在其上的一层。
感觉就是(或者还不如 MPLS with centralized TE.
另外b4的前景如何有待观察。如果google只是自己用，升级换代研发测试开销巨大，
跟其它解决方案比起来不一定有优
,
.
Internet
【在 s*****g 的大作中提到】
: SDN/OF in multi-tenant Data Center environment promises rapid/scalable
: network deployment. I don't see how much SDN/OF can help in campus/LAN
: environment, I don't think cost reduction is a big selling point of SDN/OF,
: it is more about the problems SDN/OF can solve.
: What I am really interested in SDN/OF is its potential in WAN networking,
: traditional networking separates control plan and forwarding plan only at
: device level, control plan does not even how forwarding plan is doing (i.e.
: link utilization) on a particular device, let along end to end how
: forwarding plan is behaving, this makes the applications across WAN/Internet
: suffer. Current solution for the above problem is to manul traffic-
|
d****i
发帖数: 1038 | 32 谢楼主发的第一个连接。 只有一个感想，总算见到活的了。
文章非常具体的讲了SDN的实现，回答了我以前一直不太确定的事情，很有启发性
也可以看出在未来几年内sdn对传统网络和网络设备的冲击是有限的。
cisco 的one如果作好也可以做不错的过渡。
【在 s*****g 的大作中提到】
: http://cseweb.ucsd.edu/~vahdat/papers/b4-sigcomm13.pdf
: http://www.networking-forum.com/viewtopic.php?f=46&t=29803
|
d****i
发帖数: 1038 | 33 Re this.
I think the real selling point of OF and SDN is the manageability it bring
to table. I don't mean the traditional network management here, which
actually is too focused on device management.
The 大棋 SDN and OF brings in is the focus of network management on the real
network level/end to end level and even at per flow level. And the
management includes all aspect of the network from data plane to control
plane to management plane.
but from this article it actually shows even google doesn't have a general
solution for such a 大棋. Its SDN can only solve or partly solve its own
problem.
,
.
Internet
【在 s*****g 的大作中提到】
: SDN/OF in multi-tenant Data Center environment promises rapid/scalable
: network deployment. I don't see how much SDN/OF can help in campus/LAN
: environment, I don't think cost reduction is a big selling point of SDN/OF,
: it is more about the problems SDN/OF can solve.
: What I am really interested in SDN/OF is its potential in WAN networking,
: traditional networking separates control plan and forwarding plan only at
: device level, control plan does not even how forwarding plan is doing (i.e.
: link utilization) on a particular device, let along end to end how
: forwarding plan is behaving, this makes the applications across WAN/Internet
: suffer. Current solution for the above problem is to manul traffic-
|
s*****g
发帖数: 1055 | 34 I think Cisco ONE's approach is different from what Google is doing, Cisco
ONE is more focus on providing APIs for network devices configuration, it
does not provide a central traffic control system that can take application
requirements and translate them to network device forwarding table change.
I also don't think this new idea of networking would benefit most medium to
small networks, but ISPs should take a close look at it, as they will be
able to increase their expensive long haul links' utilization and at the
same time provide VPN-like Diff services to customers.
I don't understand how the out-of-band network is built for Openflow across
WAN, I guess it still depends on in-band network built by traditional
networking, thought?
【在 d****i 的大作中提到】
: 谢楼主发的第一个连接。 只有一个感想，总算见到活的了。
: 文章非常具体的讲了SDN的实现，回答了我以前一直不太确定的事情，很有启发性
: 也可以看出在未来几年内sdn对传统网络和网络设备的冲击是有限的。
: cisco 的one如果作好也可以做不错的过渡。
|
z**r
发帖数: 17771 | 35 openflow is equivalent to just one part of the Cisco ONE, API is for
northbound which is callef ONEPK.
application
to
across
【在 s*****g 的大作中提到】
: I think Cisco ONE's approach is different from what Google is doing, Cisco
: ONE is more focus on providing APIs for network devices configuration, it
: does not provide a central traffic control system that can take application
: requirements and translate them to network device forwarding table change.
: I also don't think this new idea of networking would benefit most medium to
: small networks, but ISPs should take a close look at it, as they will be
: able to increase their expensive long haul links' utilization and at the
: same time provide VPN-like Diff services to customers.
: I don't understand how the out-of-band network is built for Openflow across
: WAN, I guess it still depends on in-band network built by traditional
|
d****i
发帖数: 1038 | 36 today most vendors are still focusing on implementing the agent part of SDN.
Cisco ONE is same, as a counterpart of OFA, to provide more flexible
programmable way for current IOS/NXOS based cisco devices.
I am not sure whether cisco plans or is doing something at controller level
though I think the big money of SDN actually will be on the controller side/
network OS and app side.
application
to
across
【在 s*****g 的大作中提到】
: I think Cisco ONE's approach is different from what Google is doing, Cisco
: ONE is more focus on providing APIs for network devices configuration, it
: does not provide a central traffic control system that can take application
: requirements and translate them to network device forwarding table change.
: I also don't think this new idea of networking would benefit most medium to
: small networks, but ISPs should take a close look at it, as they will be
: able to increase their expensive long haul links' utilization and at the
: same time provide VPN-like Diff services to customers.
: I don't understand how the out-of-band network is built for Openflow across
: WAN, I guess it still depends on in-band network built by traditional
|
v**n
发帖数: 951 | 37 OnePK中的DPSS PSS RSS都是为了你说的这些东东而存在的，不光是device control。
目前还不是很好用，不同platform上也有细微差别（搞的我的一个project很郁闷因为
要对还几个系列的设备做flow level的control）。希望以后能好一些。
application
to
across
【在 s*****g 的大作中提到】
: I think Cisco ONE's approach is different from what Google is doing, Cisco
: ONE is more focus on providing APIs for network devices configuration, it
: does not provide a central traffic control system that can take application
: requirements and translate them to network device forwarding table change.
: I also don't think this new idea of networking would benefit most medium to
: small networks, but ISPs should take a close look at it, as they will be
: able to increase their expensive long haul links' utilization and at the
: same time provide VPN-like Diff services to customers.
: I don't understand how the out-of-band network is built for Openflow across
: WAN, I guess it still depends on in-band network built by traditional
|
s******v
发帖数: 4495 | 38 多谢sharing
看了看这两个link，这个google 10ge switch，用的是24颗broadcom 56820, 16x10GE,
做的128x10G non-blocking switch. 上面运行的是linux，openflow agent as an
application level process.
第二个link里面应该是早期的产品，或者是低端的top-of-rack edge switch.
还有几个科普文章
http://www.wired.com/wiredenterprise/2012/04/going-with-the-flo
http://www.nyquistcapital.com/2007/11/16/googles-secret-10gbe-s
具体从google deployment来看，还是有些局限性
1) 只有全球几个Data Center，所以fwd table比较小，3000-4000 tunnel
2) application/service很简单，就是tunnel Fwd Group + TE, backbone网络还是比
较简单
3) 具体实现上，critical hello packet和control plane pkt还在一个Queue里面
不过这个趋势的确是不可忽视，如果这个OFA普及起来，对传统的网络设备商影响很大
【在 s*****g 的大作中提到】
: http://cseweb.ucsd.edu/~vahdat/papers/b4-sigcomm13.pdf
: http://www.networking-forum.com/viewtopic.php?f=46&t=29803
|
