L******t
发帖数: 1985 | 1 The main reason to use L2 DCN is because of VM mobility. Now with VXLAN, VM
mobility is supported as well, but this time over L3 DCN. In L3 DCN,
multipath is no longer a problem. And VXLAN can also fix the MAC scalability
issue in similar way as FabricPath. And VXLAN is even better because it
improves vlan scalability.
So in my understanding VXLAN makes FabricPath/TRILL much less appealing than
before. |
|
z**r
发帖数: 17771 | 2 vxlan 和otv 相似吗？不太了解vxlan 。 |
|
z**r
发帖数: 17771 | 3 otv is for DCI, can VxLAN do the same thing? I briefly read some ppt about
VxLAN,seems like a remote virtual interface within data center? |
|
f*****m
发帖数: 416 | 4 VXLAN/NVGRE/STT 就是竞争关系.
当然, VXLAN和OTV也是有血缘关系的 :-) |
|
w*f
发帖数: 111 | 5 OTV is better. VXlan needs a gateway which creates tunnel |
|
z**r
发帖数: 17771 | 6 otv is for site to site l2 DCI, read a little about VxLAN, but the 16M VLAN
scalability doesn't seem to be used for DCI? |
|
f*****m
发帖数: 416 | 7 OTV是DCI, VXLAN是intra-dc. |
|
|
|
|
|
L******t
发帖数: 1985 | 12 本来觉得L2MP挺有前途的，不管是FabricPath还是TRILL。但现在VXLAN好像很热，大家
都在支持，这个应该没有vendor lock-in的concern。所以又有点不看好L2MP了。谁能
给分析一下？ |
|
z**r
发帖数: 17771 | 13 VxLAN is for intra DC, not inter DC
VM
scalability
than |
|
I********x
发帖数: 858 | 14 我看了下VTS：
Open Standards
Standards-based protocols: Supports Border Gateway Protocol-Ethernet VPN (
BGP-EVPN), Virtual Extensible LAN (VXLAN), and Multiprotocol Label Switching
(MPLS)
REST APIs: Supports transparent integration with a variety of domain
managers and virtual infrastructure managers (VIMs) such as OpenStack and
vCenter
Programmability: Industry-leading, multivendor device programmability with
support for multiple hypervisors
看起来VTS只是个框架，低层实现还可以是bgp evpn？因为cisco文档在ACI内部实现的
细节很少，所以我还不... 阅读全帖 |
|
I********x
发帖数: 858 | 15 我看了下VTS：
Open Standards
Standards-based protocols: Supports Border Gateway Protocol-Ethernet VPN (
BGP-EVPN), Virtual Extensible LAN (VXLAN), and Multiprotocol Label Switching
(MPLS)
REST APIs: Supports transparent integration with a variety of domain
managers and virtual infrastructure managers (VIMs) such as OpenStack and
vCenter
Programmability: Industry-leading, multivendor device programmability with
support for multiple hypervisors
看起来VTS只是个框架，低层实现还可以是bgp evpn？因为cisco文档在ACI内部实现的
细节很少，所以我还不... 阅读全帖 |
|
f*****m
发帖数: 416 | 16 从包的封装而言, vxlan和 otv更接近.但control plane不一样.用途不一样. otv是DCI
, nvgre/vxlan都是intra DC |
|
s****t
发帖数: 698 | 17
为啥说VXLAN是intra-DC呢？
L2 over L3， VNI被写在VXLAN header里吗？
BTW：从这一点上来看，intra-DC和inter-DC貌似没有区别啊。 |
|
L******t
发帖数: 1985 | 18 It's IP overlay, like VXLAN/NVGRE/STT. And most likely VXLAN will win out at
least for extended L2.
GRE/ |
|
c*********g
发帖数: 14 | 19 We have a few openings at our San Jose office. If you are interested (or
your friends may be interested), please feel free to send your resumes to :
c**********[email protected]
Thank you very much
Company: MediaTek USA, Inc
Job position: multiple ASIC architecture, system modeling, Design engineer(
DE) engineering positions available from mid to senior level.
Job titles: Staff Engineer, Technical Manager.
Job Requirements:
- Minimum 5 years of ASIC Design experience with BSEE or MSEE de... 阅读全帖 |
|
l****u
发帖数: 95 | 20 new grads 也行，站内联系，多谢！
公司HQ在上海，成立于2000年，湾区有R&D中心。
////
We are hiring software engineer position at SIHUA Technologies in Silicon
Valley focusing on cutting-edge SDN, Container and OpenStack networking. In
this team, you will have opportunity to architect, design and develop
prototypes for proof of concept. The candidate will have to be a self-
motivated and passionate professional and you will be interacting and
working closely with Cloud platform team to deliver high value hyper-
converged IT ar... 阅读全帖 |
|
|
a**********k
发帖数: 1953 | 22 前几天开会时听了一个有关SDN的讲座，其中
提到了一大堆新的Network Virtualization
Overlay技术： 包括VxLAN, NVGRE， STT等。
大家对这些有什么想法？ 个人觉得这些
比起C家的LISP和 VN-tag Overlay技术要
简便。 |
|
f*****m
发帖数: 416 | 23 你把 vxlan的format和lisp的对比一下 |
|
a**********k
发帖数: 1953 | 24 vxlan is essentially tunneling L2 packets in UDP flow,
with a 24-bit VNI. lisp does more. Both are complementary
in some sense. |
|
L******t
发帖数: 1985 | 25 两个很相似。但应用场合的区别是什么？各自的优缺点？有好的link也行。 |
|
L******t
发帖数: 1985 | 26 They are both technologies to provide L2 connectivity over L3 IP network. |
|
d****i
发帖数: 1038 | 27 what is the difference between otv and l2vpn? |
|
L******t
发帖数: 1985 | 28 OTV and l2vpn are pretty much interchangeable usage wise. L2vpn is usually
VPLS based on MPLS infra (even though theoretically VPLS does not have to),
but OTV is only based on vanilla IP infra. |
|
z**r
发帖数: 17771 | 29 俺觉得这个不是关键，关键是L2VPN是PE上实现的FEATURE，而OTV是在CE端实现。可以
让CE不依赖于PE的限制来搞自己的东西
, |
|
z**r
发帖数: 17771 | 30 btw, there are some good slides in networkers 2012, haven't got a chance to
read much tho |
|
|
a**********k
发帖数: 1953 | 32 I switched to Chrome on Win7 to read mitbbs and experienced
various problems. Here is another one: a paragraph I posted
last night got completely lost. |
|
r********t
发帖数: 338 | 33 咱指的是ss7。
就象你说的control vs distributed 各有千秋。对PSTN来说，provision的目标明确，
就是一个连接。
provision对Internet，可就不好说了，flow, qos, acl, vpn, lan, vxlan, mpls lsp
...
provision的目标都不是很明确，central control 咋整？
design
this |
|
z**r
发帖数: 17771 | 34 does VxLAN scale the MAC?
at |
|
z**r
发帖数: 17771 | 35 in FP, there are still many limitations as for conversational MAC learning,
say, the CE VLAN still needs STP, and whenever you bring up an SVI,
conversational MAC learning is broken on that leaf ...
Does VxLAN work better? |
|
L******t
发帖数: 1985 | 36 As to interaction with STP, VxLAN is pretty much the same as FP, right?
I don't know why conversational learning is broken upon bringing up an SVI.
Is it an implementation issue, or a more fundamental issue?
, |
|
z**r
发帖数: 17771 | 37 这俩没有冲突吧？一个是解决L2 CONVERGENCE加ECMP加MAC SCALABILITY，另外一个解
决VLAN SCALABILITY |
|
L******t
发帖数: 1985 | 38 So is FabricPath/TRILL, right? |
|
s*****g
发帖数: 1055 | 39 Of course, it can do L2, L3 even VXLAN.
Spine-leaf is not exactly mesh, depending on your specific requirement (
scalability, over-subscription ratio etc), you can have multiple layers of
spine-leaf. You can have two ToRs on each rack, or run routing protocols
from servers to two ToRs in different racks, there is really nothing fancy
there. In my case, we simply do not care about a rack going off line because
application takes care of it.
No, the metric collection tools are all open source, goog... 阅读全帖 |
|
s*****g
发帖数: 1055 | 40 Current Internet Exchange (or any other business exchanges that need to
provide ad-hoc peering services) architecture is mostly composed to big
chassis routers/switches, those big boxes are configured to provide a big L2
-fabric, either pure L2 or different flavors of MPLS L2 VPNs, problem with
this design is, per-port cost of big switches are high, network can get
overly complicated in control plane, big boxes tend to less stable than
fixed configuration because the software is much more compli... 阅读全帖 |
|
s*****g
发帖数: 1055 | 41 Current Internet Exchange (or any other business exchanges that need to
provide ad-hoc peering services) architecture is mostly composed to big
chassis routers/switches, those big boxes are configured to provide a big L2
-fabric, either pure L2 or different flavors of MPLS L2 VPNs, problem with
this design is, per-port cost of big switches are high, network can get
overly complicated in control plane, big boxes tend to less stable than
fixed configuration because the software is much more compli... 阅读全帖 |
|
I********x
发帖数: 858 | 42 对的， aci本质上和J的解决方案一样，都是vxlan加bgp evpn，实现细节上有些不同。
aci实际上把部分设计内容包装进去了，比较适合企业用户不需要太多底层细节。
j的话基本都要自己设计诸如ASN，link IP等等，contrail基本是个纯overlay解决方案
，还是不放心全部网络feature都跑在hypervisor上。 |
|
I********x
发帖数: 858 | 43 对的， aci本质上和J的解决方案一样，都是vxlan加bgp evpn，实现细节上有些不同。
aci实际上把部分设计内容包装进去了，比较适合企业用户不需要太多底层细节。
j的话基本都要自己设计诸如ASN，link IP等等，contrail基本是个纯overlay解决方案
，还是不放心全部网络feature都跑在hypervisor上。 |
|
I********x
发帖数: 858 | 44 更新下：
漫长的rfq之后我们选择了Aci基于：
1. 在中小型dc情况下思科竟然有了价格优势，很大部分得益于免费的avs
2. 服务器小组比较喜欢aci的多用户模式 contrail比较专注于网路这一块 多用户要靠
高层orchestration.
3. 网路underlay这块J要在mx上做集中路由 感觉不如思科Symmetrical irb 好看 性能
方面各说各的 都不相信不做参考
4 aci的service chaining做得好些 j的解决方案还主要靠网工调路由...
实在不满j software vtep uses GRE while hardware vtep uses vxlan.
Switching |
|
I********x
发帖数: 858 | 45 思科avs nexus1000v就是为这个而生的啊
你能vlan到vm guest，其它的要firewall firewall 要iPs ips。
AVS多个vxlan功能，其它一样。contail结构差不多不过根据ppt功能要强些。
vmware只支持自己家hypervisor，如果只用vmware也不是问题。 |
|
g*****n
发帖数: 239 | 46 都是bare metal，有gateway，请问有什么产品可以实现在一个Vlan里overlay多个
vxlan？除了gateway，switch等都不受我们控制。 |
|
g*****n
发帖数: 239 | 47 都是bare metal，有gateway，请问有什么产品可以实现在一个Vlan里overlay多个
vxlan？除了gateway，switch等都不受我们控制。 |
|
I********x
发帖数: 858 | 48 你可以在cloud上跑虚拟机然后跑软交换机，如果支持vxlan cloud底层看到的都是IP，
应该没问题。
为什要这样我就不知道了。 |
|
s*****g
发帖数: 1055 | 49 Spine-leaf 显然是DC不二的选择，cheap, flexible, scalable, no vendor lock in.
至于L2 还是L3 还要看你application, scalability 的需求，如果只有少量的rack,
spanning tree works just fine, L3 to TOR 应该是标配，需要L2 inside D.C. 话可
以用evpn/vxlan 实现，trill/qfabric 应该不是大势所趋。 |
|
s*****g
发帖数: 1055 | 50 Spine-leaf 显然是DC不二的选择，cheap, flexible, scalable, no vendor lock in.
至于L2 还是L3 还要看你application, scalability 的需求，如果只有少量的rack,
spanning tree works just fine, L3 to TOR 应该是标配，需要L2 inside D.C. 话可
以用evpn/vxlan 实现，trill/qfabric 应该不是大势所趋。
Buffer size? there are two totally opposite opinions wrt why deep buffer
matter/does not matter ... it is beyond my pay grade. |
|
