Failover in the Network Fabric?

[kenlyle]

When I search Google, I find : "Results 1 - 10 of about 689,000 for failover switch"

So, in conclude that it's likely that there are network switches which can implement failover, of *something*. This may mean bandwidth sources/connectivity, I don't know.

But is there anything in the network infrastructure that can do failover of servers and/or services?

My point is that if Cisco, et al. have already invented this, it doesn't need to be re-invented, or implemented at the PHP level, etc.- having it in "hardware" should be more reliable, no?

So, can the switches, etc. at Colo redirect requests to my Apache server if it's found to be down? And, perhaps more interesting, fail back?

Best,
K

[Lloyd]

I think DNS failover implemented via DDNS and monitoring scripts running on a remote server is efficient. This is a software solution, and it works fine for me.

I think your trouble with software DNS failover (as you mentioned in an earlier post) may be due to not setting a short TTL for the A record for the domain. The TTL determines how long the non-authoritative DNS servers will "remember" the value (IP number) associated with the record.

Suppose your primary server fails, and the machine monitoring it detects that in a couple of minutes and makes the change in the DNS record to point to the backup website. That is the DNS failover action, but the non-authoritative nameservers around the world will still return the OLD (incorrect) IP for the domain until the information "expires." The expire time is determined by the TTL value.

A good TTL value for domains protected by DNS failover is, in my opinion, about 3 minutes. With this setting, about 3 minutes after your primary server fails, all non-authoritative nameservers will be obliged to re-query the authoritative nameserver for the A (IP) record. At that point, any DNS queries will return the new (failover) IP.

In sum, if the primary server fails, the secondary (failover) server takes over transparently 3 minutes after the monitoring server changes the DNS record.

I have not looked at using Pingability for DNS failover. I use a second Rimuhosting VPS, plus cron, my own perl script, nupdate, and bind9.

(I think it would be good business for Rimuhosting to offer free DNS failover, to encourage customers to have a second VPS.)

[kenlyle]

I've since implemented failover with 2 VPS' on separate sides of the Dallas Data Center, triggered by Pingability.

Thanks for the concurrence on Failover Plans as a product...it's something that I have encouraged Rimu to have a serious look at, based on the pain I had in implementing mine.

Best,
K

[Lloyd]

Glad you got it working OK.

I didn't like the idea of having my failover VPS at the same data center as the primary server. I have my primary server at Dallas and the secondary (failover/backup) server at Rimuhosting's UK facility.

[kenlyle]

In conversation with the operators of the DC in Dallas, it's effectively two DCs. Patrick maintains that even if a plane were to take out one of the buildings, the other would keep functioning, due to the redundancies that are in place.

My concern was that I didn't want to run up a big bandwidth bill, and I still believe in the concept of "netiquette" from the 90s, like using the nearest mirror site, and NOT doing trans-oceanic data transfers.

Best,
K

[Lloyd]

Hmmm, that sounds interesting (independence of the 2 Dallas DC buildings). If I had know that, I might have done something similar for my second VPS.

[kenlyle]

Yeah, I thought so. That's why I offered the info. I found it very interesting.

You can kind of see it for yourself. Go to Colo4Dallas.com, and get their Address, then use a mapping service to look at the satellite picture.

For the rest, I am just going on what Patrick told me, stuff about independent generators with separate fuel supplies, etc.