m**D
发帖数: 134 | 1 I feel the multicast address should never appear as source address. But I
met an interview question about that... |
c*****i
发帖数: 631 | 2 it's possible. if you have two servers want to hear the same reply. I
remember the MS windows load balance use multicast address as source address
. |
L******t
发帖数: 1985 | 3 I don't think so.
What's the interview question??
【在 m**D 的大作中提到】
: I feel the multicast address should never appear as source address. But I
: met an interview question about that...
|
z**r
发帖数: 17771 | 4 hmm... I agree with you. what was the question?
【在 m**D 的大作中提到】
: I feel the multicast address should never appear as source address. But I
: met an interview question about that...
|
b******e
发帖数: 66 | 5 Of course it is possible, in the old token ring/source routing days, you
will see multicast source MAC in a switch.
Anybody who asks this interview question is stupid.
【在 L******t 的大作中提到】
: I don't think so.
: What's the interview question??
|
z**r
发帖数: 17771 | 6 the load balancer should redirect only, should not do the packet replication
. can you give more details?
address
【在 c*****i 的大作中提到】
: it's possible. if you have two servers want to hear the same reply. I
: remember the MS windows load balance use multicast address as source address
: .
|
z**r
发帖数: 17771 | 7 more heads up?
【在 b******e 的大作中提到】
: Of course it is possible, in the old token ring/source routing days, you
: will see multicast source MAC in a switch.
: Anybody who asks this interview question is stupid.
|
c*a
发帖数: 806 | 8 this reminds me of stonebeat (checkpoint cluster), believe multicast mac
address is being associated with unicast IP, thus two servers can have same
copy of traffic. Old Cisco IOS code did not support multicast mac in arp, so
a static arp entry is necessary in this case
address
【在 c*****i 的大作中提到】
: it's possible. if you have two servers want to hear the same reply. I
: remember the MS windows load balance use multicast address as source address
: .
|
c*a
发帖数: 806 | 9 token ring doesn't have same multicast mac as ethernet, I think it only has
group mac address. anyway, ops was asking ethernet particuarly
【在 b******e 的大作中提到】
: Of course it is possible, in the old token ring/source routing days, you
: will see multicast source MAC in a switch.
: Anybody who asks this interview question is stupid.
|
b******e
发帖数: 66 | 10 In TR networking the G/I bit is set in source MAC to indicate an source
routing frame, the bit is called RII (routing information identifier), so
the MAC address looks like a multicast MAC but it has nothing to do with
multicast.
Back to the original question, A ethernet switch may switch a frame with
multicast source, why not? it should only make switch decision based to
destination MAC.
【在 z**r 的大作中提到】
: more heads up?
|
|
|
L******t
发帖数: 1985 | 11 Don't forget switch is doing MAC learning along the way.
【在 b******e 的大作中提到】
: In TR networking the G/I bit is set in source MAC to indicate an source
: routing frame, the bit is called RII (routing information identifier), so
: the MAC address looks like a multicast MAC but it has nothing to do with
: multicast.
: Back to the original question, A ethernet switch may switch a frame with
: multicast source, why not? it should only make switch decision based to
: destination MAC.
|
L******t
发帖数: 1985 | 12 One thing is fore sure, it's a stupid question at best. Whoever asked it
doesn't know much about networking.
【在 b******e 的大作中提到】
: Of course it is possible, in the old token ring/source routing days, you
: will see multicast source MAC in a switch.
: Anybody who asks this interview question is stupid.
|
c*****i
发帖数: 631 | 13 in windows load balance, there is no hardware load blancer in front of the
servers. you need configure the two servers to give them a virtual ip
address. so in order for all servers to receive traffic, u need use
multicast address. if u use unicast address, the router will only send
traffic to one of the servers. although this design is not that good, but if
you dont have a hardware load balancer in front, that's the only choice. |
c*****i
发帖数: 631 | |
c*a
发帖数: 806 | 15 you sure it uses mcast mac in source field? the return traffic from
individual server should still use BIA as source mac
can't find the reference though
if
【在 c*****i 的大作中提到】
: in windows load balance, there is no hardware load blancer in front of the
: servers. you need configure the two servers to give them a virtual ip
: address. so in order for all servers to receive traffic, u need use
: multicast address. if u use unicast address, the router will only send
: traffic to one of the servers. although this design is not that good, but if
: you dont have a hardware load balancer in front, that's the only choice.
|
z**r
发帖数: 17771 | 16 if it's load balancing, why the router needs to send the traffic to all the
servers? if it's a cluster, the router should send the traffic to the active
server.
just curious how it works
if
【在 c*****i 的大作中提到】
: in windows load balance, there is no hardware load blancer in front of the
: servers. you need configure the two servers to give them a virtual ip
: address. so in order for all servers to receive traffic, u need use
: multicast address. if u use unicast address, the router will only send
: traffic to one of the servers. although this design is not that good, but if
: you dont have a hardware load balancer in front, that's the only choice.
|
z**r
发帖数: 17771 | 17 for www post? don't know
【在 c*****i 的大作中提到】
: 另外怎么引用别人的发言，弱弱的问一句
|
c*a
发帖数: 806 | 18 stateful failover
some f/w keep separate links between two member servers for both state
synchronization and heartbeat.then no need to send duplicate traffic
the
active
【在 z**r 的大作中提到】
: if it's load balancing, why the router needs to send the traffic to all the
: servers? if it's a cluster, the router should send the traffic to the active
: server.
: just curious how it works
:
: if
|
c*****i
发帖数: 631 | 19 two server are active and service user at the same time with a single ip
address. you can take a look the microsoft doc. it states quite clearly. |
m**t
发帖数: 1292 | 20 Saw the stuff before, there is a filtering layer built below the IP stack on
each node, all the nodes share a same virtual IP and MAC address. It is not
popular method but it is do-able. The advantage is that the cluster does
not need a seperate load balancer to do the load balancing but rather push
the function to individual nodes, it is also able to provide service to IP
level apps such as IPsec, MIP etc so those IP layer addon can potentially
benefit from seeing all the communication states a
【在 z**r 的大作中提到】
: if it's load balancing, why the router needs to send the traffic to all the
: servers? if it's a cluster, the router should send the traffic to the active
: server.
: just curious how it works
:
: if
|
|
|
z**r
发帖数: 17771 | 21 把自身的关键HA应用建立在别人的multicast转发？有点奇怪
【在 c*a 的大作中提到】
: stateful failover
: some f/w keep separate links between two member servers for both state
: synchronization and heartbeat.then no need to send duplicate traffic
:
: the
: active
|
z**r
发帖数: 17771 | 22 link?
【在 c*****i 的大作中提到】
: two server are active and service user at the same time with a single ip
: address. you can take a look the microsoft doc. it states quite clearly.
|
z**r
发帖数: 17771 | 23 早期的HACMP都是这样，当时也没有load balancer。都是一个cluster share一个
virtual IP，现在这样的应用比较少，但是还是有，solaris, linux等都可以。但是俺
做过tcpdump，
没有看到过active server发multicast啊。不过俺没做过两个都是active的情况
on
not
state
【在 m**t 的大作中提到】
: Saw the stuff before, there is a filtering layer built below the IP stack on
: each node, all the nodes share a same virtual IP and MAC address. It is not
: popular method but it is do-able. The advantage is that the cluster does
: not need a seperate load balancer to do the load balancing but rather push
: the function to individual nodes, it is also able to provide service to IP
: level apps such as IPsec, MIP etc so those IP layer addon can potentially
: benefit from seeing all the communication states a
|
