Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 6 Aug 2000 16:37:14 +0200
From:      Gabriel Ambuehl <gabriel_ambuehl@buz.ch>
To:        "David TOUITOU" <david@solexine.fr>
Cc:        freebsd-questions@freebsd.org
Subject:   Re[6]: Network load balancing hardware ...
Message-ID:  <95106139430.20000806163714@buz.ch>
In-Reply-To: <398D9042.7246.2DFDBD87@localhost>
References:  <398D514E.27534.2D07D590@localhost> <398D9042.7246.2DFDBD87@localhost>

next in thread | previous in thread | raw e-mail | index | archive | help
Hello David,

Sunday, August 06, 2000, 4:20:18 PM, you wrote:
>> Outgoing traffic won't be a problem in any case as the backend servers
>> could be configured to use another router. 
> I don't get your point here.

Well, configuring our servers to use another router (which has got
another IP) in case the primary one fails is rather easy.
Getting the clients to use it is the big problem.

>> We've recently
>> had a box in a state where it did respond to pings but no daemons were
>> responding for about two hours... 
> Eddie doesn't rely on "ping" but on daemons tests.

I know. But what's how do you want to do an IP takeover if the normal
box still holds that IP? Having two identical IPs in a network isn't
something one would want (in most cases, TCP/IP just screws up rather
completely but of course that's depending on the involved machines).

> There is one Eddie daemon (or similar) running on each box (DNS, 
> frontend and backend) and they all communicate wichi each other.

Clear.

> That's the way the DNS sends the request to the "correct" frontend 
> server (with rules such as geographic load balancing, local laod 
> balancing or failover) and to the "correct" backend server (same 
> thing: you can parameter the DNS server to send 10% of the request on 
> one server, 30% on antoher one and the rest to the third one, and so 
> on).

If you've got enough IPs, this is even possible with vanilla Bind
(though not with geographic load balancing): give the faster box three IPs,
the slower one just one and then use the normal round robin features.

>> I for my self can see no way to
>> protect you from such a scenario without some hardware failover
>> technologies. 
> I've had a look to hardware load balancers and failover (such as
> alteon or Cisco's LocalDirector) : they are "huge" single point of 
> failure.

Some of them have got integrated devices which enable you to cross
connect two of them for failover. OTOH in this field, one should first
get "failsafe" routers, I think.

> Eddie doesn't rely on one only box, that's the main idea.

I know and I consider this to be good. But I'm still not convinced that real failover is possible
without any additional hardware. You NEED a facility to take down a
server which is out of control. Just relying on some software
shutdown mechanism won't be sufficient when major trouble starts (as
already mentioned, there ARE situations when a box has got an IP but
doesn't do ANYTHING). Getting the backup machine in service is easy enough.
A safe way to get the failing primary COMPLETELY out of service
is a bit harder (the easiest but also the riskiest way of doing
it would be a hardware reset for the primary).



> I'll double check Eddie's website to try to find out more info about 
> this.

I've already tryed to gather some info about this topic on the EddieML
but that one is very silent...




Best regards,
 Gabriel




To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?95106139430.20000806163714>