thanx for quick responses. The easiest solution seems to be the round robin
DNS. Thanx to all responder. The best description cames from Leif Hedstrom
>Dirk Boenning writes:
>Cause of availability and the extremly high numbers of parallel =
>to the Webserver I want to spread one Webservice over at least two ultra =
This is traditionally done using what's called DNS round-robin.
Basically you'll configure your DNS with (probably) a CNAME for
www.a.com looking like:
www.a.com 1800 IN CNAME www1.a.com
www.a.com 1800 IN CNAME www2.a.com
www.a.com 1800 IN CNAME www3.a.com
This will give you a reasonable load balancing, but it's not perfect
(since DNS is caching). The number 1800 tells other DNS servers to keep
your entries for half an hour. On average you should be getting pretty
fair distribution, specially if you have lots of customers.
If machine www2.a.com goes down permanently, you just have to change
your DNS server to remove it from www.a.com DNS RR.
Since you are fortunate to use Solaris2.x, you can even have one of the
other machines take over www2.a.com's IP number with something like this
ifconfig hme0:1 inet www2.a.com up
(This means www1.a.com will be known on the network as both www1 and www2.
When www2 is back again, do "ifconfig hme0:1 down" on www1.)
This way, even if people try to go to the broken machine (www2.a.com),
they will get a response. This is important to remember, since even
after you change DNS, it will take time until all caching DNS servers
update their entries. Also remember that Netscape Navigator does it's
own DNS caching.
Now, If you are really tricky, invent your own "protocol", and make
machines "back up" each other automagically (it's been done before...).
So when any of the www machines goes down, one of the other takes on his
lost comrade's duties! :)
I hope this helps.
------------- End Forwarded Message -------------
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:03 CDT