s****l
发帖数: 16457 | 1 我看的是英文解说。为啥解说一直说日本“ingression”而不是“invasion”呢？这不
是和日本教科书一致了么？ |
|
n****g
发帖数: 14743 | 2 网络工程师网夏晨曦爆料警如何监控P民zz
http://canyu.org/n83022c6.aspx
我叫夏晨曦，是中国一家运营商（ISP）的核心网工程师，在两年的职业生涯中，多次
与网警打过交道，了解他们的运作方式。在2012年10月15日的下午2点，上海市公安局
闸北分局的网警来到我们公司，要求在他们管辖范围内的一个机房（万荣一路工业园）
安装监控服务器，我被领导安排配合他们做技术支持。实际上我们公司的所有机房都有
公安的监控服务器。
我们公司的网络拓扑（部分）大致如下：以万荣路机房为例。
华为9306 三层交换机 （万荣路机房）
|
|
GW6700 OLT （万荣路机房）
|
|
GT810onu （用户家）
GW6700
GW6700 OLT（optical line terminal）共有4个上联口，第一上联口接入华为9306三层
交换机，第二个上联口接网警的监控服务器，我所提供的技术支持就是把第二个接口设
置为镜像端口，然后关联到第一个接口。GW67... 阅读全帖 |
|
t***u
发帖数: 20182 | 3 I am standing on the threshold about to enter a room. It is a complicated
business. In the first place I must shove against an atmosphere pressing
with a force of fourteen pounds on every square inch of my body. I must make
sure of landing on a plank travelling at twenty miles a second round the
sun - a fraction of a second too early or too late, the plank would be miles
away. I must do this whilst hanging from a round planet head outward into
space, and with a wind of aether blowing at no one k... 阅读全帖 |
|
|
c****t
发帖数: 19049 | 5 THE PACIFIC JOURNAL OF ADAM EWING
Thursday, 7th November—
Beyond the Indian hamlet, upon a forlorn strand, I happened on a trail of
recent footprints.
Through rotting kelp, sea cocoa-nuts & bamboo, the tracks led me to their
maker, a White man,
his trow-zers & Pea-jacket rolled up, sporting a kempt beard & an outsized
Beaver, shoveling
& sifting the cindery sand with a teaspoon so intently that he noticed me
only after I had hailed
him from ten yards away. Thus it was, I made the acquaintance of... 阅读全帖 |
|
c****p
发帖数: 6474 | 6 出去旅游的时候看到个教堂的portal，
这个Portal的描述是：
A place to brain wash fragile minds.
看完我会心一笑。 |
|
b*****n
发帖数: 4976 | 7
----------
Brainwash 变了。brain wash
我都会心笑了一笑。
你真本事。 |
|
|
p*****h
发帖数: 3 | 9 【 以下文字转载自 CS 讨论区 】
【 原文由 peterzh 所发表 】
I am reviewing a paper.
The paper said:
Given an autonomous system including edge routers and core routers.
Each edge router knows the network topology and link state by OSPF protocol.
Now for each egress edge router a sink tree needs to be found, where
each leaf is an ingress edge router and each link in the tree satisfies
some conditions, such as bandwidth and delay constraints.
It can be proved the problem is NP-complete.
Then in reality some heuristic a |
|
p*****h
发帖数: 3 | 10 I am reviewing a paper.
The paper said:
Given an autonomous system including edge routers and core routers.
Each edge router knows the network topology and link state by OSPF protocol.
Now for each egress edge router a sink tree needs to be found, where
each leaf is an ingress edge router and each link in the tree satisfies
some conditions, such as bandwidth and delay constraints.
It can be proved the problem is NP-complete.
Then in reality some heuristic algorithms can be used to find such a
tr |
|
|
b******e
发帖数: 66 | 12
label
If you are trusting cos, exp-bit will be automatically copied from cos, you
don't need to remark exp.
How did you set the VC label exp? and what is the point? vc label is not used
for switching.
So on the exgress PE side, after the
the
Not necessarily, depends what QoS mode you want, for short pipe mode, you cos
value is retained, no matter what Exp value you had along the way. For uniform
mode, cos, tunnel exp, vc exp are the same at ingress PE, if for some reason,
tunnel label changed a |
|
b******e
发帖数: 66 | 13
label
If you are trusting cos, exp-bit will be automatically copied from cos, you
don't need to remark exp.
How did you set the VC label exp? and what is the point? vc label is not used
for switching.
So on the exgress PE side, after the
the
Not necessarily, depends what QoS mode you want, for short pipe mode, you cos
value is retained, no matter what Exp value you had along the way. For uniform
mode, cos, tunnel exp, vc exp are the same at ingress PE, if for some reason,
tunnel label changed a |
|
u*v
发帖数: 18 | 14 俺对QoS研究很少，我想至少在MPLS TE有用，对进来的Traffic进行分类，确认打什么标
签，走哪条路经，以后就是标记交换了，所以必须在进MPLS时就打上tag。 |
|
u*v
发帖数: 18 | 15 俺对QoS研究很少，我想至少在MPLS TE有用，对进来的Traffic进行分类，确认打什么标
签，走哪条路经，以后就是标记交换了，所以必须在进MPLS时就打上tag。 |
|
z**r
发帖数: 17771 | 16 这个俺同意，但是分类(classification)严格上说并不是QoS啊，分类了以后接下来干的
事情才更重要
根
加
成
客 |
|
L******t
发帖数: 1985 | 17 QoS is implemented by priority queues. That's why the earlier we classify
traffic into different queues, the better.
的 |
|
z**r
发帖数: 17771 | 18 but the problem is, if the interface/lien card is not oversubscribed, the low
latency queue is actually not meanning too much, actually, in this situation,
as long as we have enough queue depth, we even don't need to worry about the
priorities
, |
|
w********n
发帖数: 7 | 19 感觉你在某个router vendor有间谍啊
解 |
|
|
z**r
发帖数: 17771 | 21 if the interface is not oversubscribed, in most cases(if no line card
oversubscription), the queue depth would be always 0. we even don't need a
very big buffer.
the situation you mentioned is exactly what I am concerned. the low latency
queue on port B would not mean much.
interfaces
has |
|
z**r
发帖数: 17771 | 22
第一个不应该是，如果到了fabric都oversubscribed了的话，任何traffic都没有办法保
证了。除非是一个特殊的traffic pattern。
can you explain more? |
|
L******t
发帖数: 1985 | 23 我上上篇帖子没说清楚。假设一个controller/arbiter控制两个PHY port A & B；port A
满载，但全是垃圾traffic，port B undersubscribed，但是VOIP traffic。这种情况下
，如果两个PHY port分别只有一个queue的话，arbiter作round robin，port B traffic
will suffer even though it's undersubscribed. Port A的垃圾traffic会一路上去让
别人suffer直到classification point.不用说更经常的情况是一个arbiter控制4个或者
更多PHY ports.
所以说如果单个port不oversubscribed,不影响multiple queueing的必要性.如果都不
oversubscribed的话,共产主义就来到了.
个 |
|
z**r
发帖数: 17771 | 24 来自主题: EmergingNetworking版 - 反无线路由？ 路由就是路由，没有ingress/egress之分，也没有进internet/出internet之分 |
|
g**d
发帖数: 77 | 25 好像现在市面上的都是asic的，只有jnpr的新的是用np2，我也不知道是哪个系列的。
np2/np3是10g level的，而且是ingress／egress各一片，要是40g就要4片最少。新的
np4才是40g，08年才sample。
其他的high end npu还有broadcom的sunburst，xcelerate和bay system，不过他们好
像没有大的production。 |
|
L******t
发帖数: 1985 | 26 STP has nothing to do with L2 forwarding table.
STP is just used to shut down some ports to avoid loops.
Conceptually, L2 forwarding table is simply a list of (src-mac, ingress-port
) pair, stuffed by source mac learning.
At layer 3, TTL is the last straw to avoid loop. You can think the routing
protocol can guarantee loop-free forwarding. |
|
z**r
发帖数: 17771 | 27 oversubscription, right? |
|
L******t
发帖数: 1985 | 28 Is this your topology?
[Router 1]
|
--+---------+------+----
| |
[Router 2] Host(192.168.10.x)
For host, the ping should be sent to Router 2 directly, but in your case
sent to Router 1. Router 1 found that the egress IF is the same as ingress
IF, so replied an ICMP redirect pak.
You should configure static route to 192.168.20.0 on each host in 192.168.10
.0.
1 |
|
e**********w
发帖数: 117 | 29 是的. 这个方法有这个不足. juniper的fabric基本上没有什么智能,非常简单, 所以所
有的packet processing都在PFE上做了. 用binary tree还算聪明一点的做法, off
load ingress PFE的replication工作量.
cisco不同platform的实现方法也不一样. |
|
s******v
发帖数: 4495 | 30 而且layer 2怎么会有fib？ cef，还是读ingress packet dst ip，和layer 2也没关系
啊。 |
|
s*****g
发帖数: 1055 | 31 How did IOS solve 4K interface limitation on 7600? If customer needs to have
more than 4K MPLS TE ingress tunnel, say 16K, can 7600 handle that? |
|
w*f
发帖数: 111 | 32 update
answer to "What is Juniper QFabric in layman's term?"
Rakesh Singh:
Qfabic is a new network architecture from Juniper which makes the whole data
center behave as if it were one giant switch.
There are 3 componets to QFabric: Node, Interconnect and Director. To take
the analogy of a typical Chassis-Switch, the QF/Nodes are the line card in
chassis, The QF/Interconnect is the Switch Fabric and the QF/Director is the
CPU card.
The typical chassis switch is limited to 8 or 16 line cards ... 阅读全帖 |
|
|
x*********n
发帖数: 28013 | 34 coworker有个case，就是Cannot Apply "service-policy output " on
the Interface
然后呢，TAC回复
I believe the egress service policy is limited to matching based on DSCP
values, while you have more options for ingress. Queueing evaluates the
internal DSCP and determines which of the four egress queues in which to
place the packet. The DSCP value is mapped to a CoS value, which selects one
of the queues.
Please configure a new policy-map used for egress interface matching only
DSCP values and remove... 阅读全帖 |
|
t*******r
发帖数: 3271 | 35 ingress queuing & egress queuing? |
|
t*******r
发帖数: 3271 | 36 你说的是对的, 但是现在也有ingress queuing, 就是在入接口板卡去往fabric的方向
也做调度.
常见于一些增强型queuing板卡 |
|
|
b***p
发帖数: 700 | 38 1) /24是最小的单位，不可能ISP给你Advertise更细的，不然的花，BGP Table就要增
加10倍
2）你是video contents provider，下行Traffic远大于上行的，为什么要care用户
ingress route? 你做好自己的engineering进行啊，是用的TCP？担心Asymmetric
paths?
3) 你的客户如何Fail over是他们和ISP的事情，你也管不到, 你就是需要milti
homing, 两边都advertise，That's it
25 |
|
z**r
发帖数: 17771 | 39 是不能完全等同于，但是比较接近。他这个条件没有完全表达清楚，比如1-2之间是什
么link，经过多少hop，中间都有些什么设备等等，如果只是直连，而且ingress/
egress上没有什么特殊配置，1->2和2->1能有什么差别呢。所以他既然给出的是一个模
糊的题目，就不要期望答案很清楚。还有一个办法就是1和2都和同一个NTP server
sync，这样可以最大限度的保证时间同步，那么直接计算tcp timestamp的差别也不是
不可能。
其实这就是一个最典型的IP SLA命题 |
|
