Slashdot Mirror
Internet Giants Prepare for WorldCom 'Storm'
swight1701 writes: "MSNBC has an article about how E-bay and others are enacting contingency plans in case WorldCom goes under and no one steps in to run UUNet right away. Also talks extensively about how this is happening already in Europe with KPNQwest, who is telling their customers, 'During this week you can already expect outages to happen that we cannot solve any more. At the end of this week we expect that larger parts of the network will be down.' Can telecommunications giants realistically keep up with the public's need for ever-growing bandwidth without going bankrupt?"
Yes, but that's difficult without portable address space. Even if you have redundant links, if you need to push through a DNS change to activate incoming connections on the alternate line, then you are screwed for several days, unless you keep your SOA TTL very low at all times, which is inefficient.
I've had enough abrasive sigs. Kittens are cute and fuzzy.
BGP! If our UUNet link goes down, we might drop a couple of packets, but we're instantly flipped over to our backup provider.
That is if all goes well, but the truth get's weird about BGP. Last month, with the Mae-East part of Genuity going down ( for about 3 hours ) our BGP kicked in and had packet loss of about 12% to 20%, Why? Weird peering agreements.
During those 3 hours I had the joy of running traceroutes all over the nation and watching the interaction of different carriers. it was weird to watch traffic start with one carrier go on then hit genuity then jump off to another carrier and then make it to the web servers. ( sometimes from the other side of the world LOL )
-Onepoint
if you see me, smile and say hello.
I'm sure this is obvious if you think about it -- Worldcom fell apart because of a crisis of investor confidence, not bankruptcy.
the problem is address space.
/21 of usable address space. (Yes folks thats 2048 addresses), you are forced to get your address space from an upstream provider.
/21 (2048 addresses) to get a /20 (4096 addressess).
/21.
/22 /23 /21, and hopefully get a /20. It helps if you show that you are growing.
/22 but i'm opening another withing 6 months /23 but i'm opening another withing 6 months
/23 or you may get filtered by other ISPs.
. html
If you have a web site that has less than a
The rules on arin: http://www.arin.net/policy/ipv4.html#multihomed
say that you need to show that you have used the equivalent of a
With these 4096 addresses you can then have "portable addresses".
This means that you do not need to get your address space from an upstream provider.
So lets imagine these scenarios:
1: Single homed (1 connection) company.
Potential problems:
If upstream goes out of business, then you have nowhere to route to.
To migrate you have to get a BRAND new set of IPs for your company.
You better hope that your DNS TTL is low, and that all the places out there that cache it honor it and dont set it to something astronomically high. (this does occur).
2: Multi homed (N connection (n>1) company. (running BGP).
Potential Problems:
If upstream goes out of business, then you are still routing the address space of your out of business provider out your alternate provider. (good).
You are now at arin's mercy depending on what they do with the upstreams IP space. You can bet that they will not give you a small chunk to stay as who you are.
Other large providers may get crazy ideas and start filtering. Ie: Worldcom has not paid me so I filter their netblock as protest. (see PSI).
The best help in this scenario is your own problem, the small IP range you have. Since you have a small range, you then have a "more specific" route to your network. That will override most things as null routes etc. (it will not override ACLs though).
Pretty much this scenario will keep you on the net for longer, but, should the ISP you have your space from go down and stay down, you will need to migrate address space, dns ttl etc.
3: Multi homed (N connection (n>1) company at multiple datacenters.
Potential problems:
Hopefully you have different providers at each datacenter, or at least the address space is given to you by different providers.
In this case, the worst you lose is 1 datacenter. hopefully your site can maintain full traffic out of the remainder of your datacenters.
The biggest problem here is again DNS, but if you are doing multiple datacenters, you can probably remove 1 out of the picture realyl quick.
--- This does not end the list of possible scenarios, there are many others you can do. (for example: you could have address space from 2 carriers at the same datacenter, and multihome/map addresses from each carrier onto them. ) etc.
Now as to the likelyhood, when Exodus went bankrupt for a few months last year, they did not lose all their advertisments, and for the most part they did stay up. I'd guess this would be the same thing as uunet.
Also UUnet has much more traffic than exodus. that hopefully means that most ISPs will not kill their peering with uunet.
That being said, if the rats leave the ship called uunet, uunets peering will fall to low levels, and then ISPs will be able to contractually cancel with uunet. Possible, but not likely.
Now, if you wanted to have multiple carriers, and be truly independent of any of them going away, you have to show effective usage of a
The good part of this is that you can split this into multiple sections:
ie: I have 2 datacenters, and each need a
ie: I have 4 datacenters, and each need a
you can seperate address space out to that, and show use of a
ie: I have 2 datacenters, and each need a
ie: I have 4 datacenters, and each need a
What you have to look out for is that you probably should not advertise anything more than a
for example take this recent nanog posting: http://www.merit.edu/mail.archives/nanog/msg01717
Too many specific routes (/24 and above) add more work to peoples bgp routers, as such limiting accepted routes helps performance of the router, and keeps things more stable.
In summary,
if you have 1 carrier, get 2.
if you do not have enough address space, be prepared to have to change it.
If you have multiple datacenters you should be good to go, but with some exposure as you have to re-ip.
if you have enough address space to be portable, you are good to go.