Re: Does perl have Failover with Open Source Web Platforms?

2003-03-15 Thread Christian Brink
> If a 30 second delay is acceptable, even round robin dns forms a type of
>  failover. If the first ip fails, the browser tries the second, etc.

This is incorrect. IE does support some aliveness checking
(http://www.geocities.com/tufansevim/dnsroundrobin.html) and will cycle
through A records till it gets a response, but this is not specification
and most dns resolvers and will not work for all clients.
(http://www.acmebw.com/askmrdns/archive.php?category=81&question=359).

Round Robin DNS should only be used in load sharing (it does not load
balance) scenario until most DNS resolvers handle DNS SRV records, which
will be the correct way for a DNS to provide failover.
(http://www.isc.org/ml-archives/bind-users/2000/09/msg00350.html)

LVS, as you mentioned, is an excellent solution. If you want to stay with
a DNS solution then one of the Dynamic DNS servers may fit the bill, but
your end up with other problems.
- the name server propagation time you'll have
- the problem of driving a lot of traffic to your DNS server because of
the low TTL's.
- DNS servers that don't honor TTL times that are too low - then giving
out that bad info to it's users

Christian Brink







Re: Does perl have Failover with Open Source Web Platforms?

2003-03-13 Thread Perrin Harkins
Richard Heintze wrote:
He needs declarative role based authorization and
authentication for his web site -- and maybe fault
tolerance too depending on the price of the hardware
for a linux server.
These are two separate things.

Authen/Authz can be implemented any way you like on mod_perl.  It does 
not impose an application structure.  If you want an already built 
implementation, you could look at some of the many auth modules on CPAN 
or at OpenInteract which includes a user/group security model.

For the CPAN list, start here:
http://www.cpan.org/modules/00modlist.long.html#ID15_WorldWideW
and look at the PerlAuthenHandler and PerlAuthzHandler modules.
For failover, the question is what are you failing over?  Any old 
load-balancer will give you failover between machines.  Failing over 
your data is a matter of how you implement your application.  I believe 
JBoss fails over session data.  You get the same result with mod_perl if 
you store your session data in a database, using something like 
Apache::Session.

Also, what are your favorite hardware vendors for
linux clusters?
Red Hat sells a software failover system that you can use, but most 
people just use a load-balancing switch like big/ip for this.  No 
special hardware or software is required for that.

- Perrin



Re: Does perl have Failover with Open Source Web Platforms?

2003-03-13 Thread Jay Thorne
On March 13, 2003 11:48 am, Richard Heintze wrote:
> My client is partial to perl so I installed mod_perl
> on Apache HTTPD on his windows servers.
>
> Now, however, he wants to price a linux cluster with
> raid to replace his windows servers.
>
> He needs declarative role based authorization and
> authentication for his web site -- and maybe fault
> tolerance too depending on the price of the hardware
> for a linux server.
>
> JBoss is cheap and open source with these features.

And its java and may not perform as well as well formed mod_perl, but I'm 
probably opining from my impressions based on BEA on slowlaris compared to 
mod_perl on linux.

> My client, however, is partial to perl. How can we get
> perl with failover on a linux cluster?

Lots of different ways. Depends on where you want the failover;

If a 30 second delay is acceptable, even round robin dns forms a type of 
failover. If the first ip fails, the browser tries the second, etc. 

Then you separate application server from db, and put the db on some kind of 
redundancy/failover; like a app level retry list, a OS level failover, with 
proxy ARP and whatever to minimize downtime etc.

An overview of OS level failover on linux:
http://www.linuxvirtualserver.org/Documents.html

So in summary, failover is a big fat word with lots of meaning. 

The faster you want it, the more money you spend. A 30 second failover was 
good enough for me, so I used round robin dns. Your mileage my vary.

> Also, what are your favorite hardware vendors for
> linux clusters?

Whitebox hardware with name brand components, but I'm cheap.

This is another question that is too broad. What level of service do you 
require? My local clone shop knows dick about linux but is happy to sell me 
machines with no OS loaded. You may want Scyld.com or one of the other 
cluster specialists if you are talking getting vendor expertise. They tend to 
focus on compute clusters, not web.


-- 
Jay "yohimbe" Thorne  [EMAIL PROTECTED]
Mgr Sys & Tech, Userfriendly.org