hi, Jack.
You say that " The focus is to discover errors in the
new configuration before they are used on the actual network."
but I want to know
the result before the new configuration actually used but only used in your patch
and
the result when the new configuration are used on the actual network.
the two results, do you think , are same results? If your networks is not a stub AS, I think they are different.
That is to say, perhaps your model is not a evaluation of the old configuration, and not a evaluation of the new configuration in actual use.
-----ÔʼÓʼþ-----
·¢¼þÈË: Jack Y Lin [mailto:jacklin@andrew.cmu.edu]
·¢ËÍʱ¼ä: 2004Äê2ÔÂ3ÈÕ 6:44
ÊÕ¼þÈË: Pan Zhen Yu
³ËÍ: quagga-dev@lists.quagga.net
Ö÷Ìâ: Re: [quagga-dev 842] about lin's bgp patch.
Hi Pan,
When a network operator makes a change to his bgp configuration, mainly
two types of errors can occur, routing error and policy violation. My
patch helps to discover routing error, which are incorrect path
selections in the router that is updated with the new configuration. (ie,
wrong local RIB)
The focus is not to use the new configuration to evaluate the old
configuration as you have said. The
assumption we make is that the old configuration is correct; after all,
the network is running fine and customers are not complaining. So
assuming that, changes you make in the new configuration will
probably modify your local RIB in some way. The patch compares the two
RIBs (one under the old configuration and one under the new configuration) and output changes
to the RIB, so the network operator can evaluate that before putting the
new configuration into the running network.
If this is still unclear, email me again.
Cheers,
Jack
On Mon, 2 Feb 2004, Pan Zhen Yu wrote:
> I think perhaps there are some bugs in MCU Lin's bgp patch. They are not bugs of coding, but of ideas.
>
> BGP is a dynamic routing protocol, the best routes of bpg is not only decided by any deamon connect to u, but also urself. So it is useless if you want to use a blackhole deamon to simulate a real deamon even you put the dummy deamon on the right postion and with the right connections.
>
> for example.
>
> router A --------router B----router D ---------router C
> | |
> |----------------- your Router--------------------|
> |----------------- your Dummy-------------------|
>
> U can see , the router A and route C, whose decisions depend on your Router's decision, when you put your Dummy beside your Router, the decisions you receive is desicion made by all the routers, remenber, it is also include your Router( Because at least Router A and Router C are influenced by your Router). That is to say, It's a result made by not only your new configuration( Input configration of your Dummy) ,but also your old configuration(Input and output configuration of your Router). So it is useless to use the new configuration to evaluate the old configutation.
>
> Do you think so?
> >
>
You say that " The focus is to discover errors in the
new configuration before they are used on the actual network."
but I want to know
the result before the new configuration actually used but only used in your patch
and
the result when the new configuration are used on the actual network.
the two results, do you think , are same results? If your networks is not a stub AS, I think they are different.
That is to say, perhaps your model is not a evaluation of the old configuration, and not a evaluation of the new configuration in actual use.
-----ÔʼÓʼþ-----
·¢¼þÈË: Jack Y Lin [mailto:jacklin@andrew.cmu.edu]
·¢ËÍʱ¼ä: 2004Äê2ÔÂ3ÈÕ 6:44
ÊÕ¼þÈË: Pan Zhen Yu
³ËÍ: quagga-dev@lists.quagga.net
Ö÷Ìâ: Re: [quagga-dev 842] about lin's bgp patch.
Hi Pan,
When a network operator makes a change to his bgp configuration, mainly
two types of errors can occur, routing error and policy violation. My
patch helps to discover routing error, which are incorrect path
selections in the router that is updated with the new configuration. (ie,
wrong local RIB)
The focus is not to use the new configuration to evaluate the old
configuration as you have said. The
assumption we make is that the old configuration is correct; after all,
the network is running fine and customers are not complaining. So
assuming that, changes you make in the new configuration will
probably modify your local RIB in some way. The patch compares the two
RIBs (one under the old configuration and one under the new configuration) and output changes
to the RIB, so the network operator can evaluate that before putting the
new configuration into the running network.
If this is still unclear, email me again.
Cheers,
Jack
On Mon, 2 Feb 2004, Pan Zhen Yu wrote:
> I think perhaps there are some bugs in MCU Lin's bgp patch. They are not bugs of coding, but of ideas.
>
> BGP is a dynamic routing protocol, the best routes of bpg is not only decided by any deamon connect to u, but also urself. So it is useless if you want to use a blackhole deamon to simulate a real deamon even you put the dummy deamon on the right postion and with the right connections.
>
> for example.
>
> router A --------router B----router D ---------router C
> | |
> |----------------- your Router--------------------|
> |----------------- your Dummy-------------------|
>
> U can see , the router A and route C, whose decisions depend on your Router's decision, when you put your Dummy beside your Router, the decisions you receive is desicion made by all the routers, remenber, it is also include your Router( Because at least Router A and Router C are influenced by your Router). That is to say, It's a result made by not only your new configuration( Input configration of your Dummy) ,but also your old configuration(Input and output configuration of your Router). So it is useless to use the new configuration to evaluate the old configutation.
>
> Do you think so?
> >
>