Redundancy

M@tt · 26th August 2003

What are people's thoughts on best ways / cost effective ways of providing redundancy for important websites? (Redundancy accross multiple servers).

We (JPL in this case) are looking at redundancy on multiple servers in geographically diverse locations, however am not particularly familiar with what hardware / software and other things would be involved.

racksense · 26th August 2003

Quote:

Originally Posted by M@tt

What are people's thoughts on best ways / cost effective ways of providing redundancy for important websites? (Redundancy accross multiple servers).

We (JPL in this case) are looking at redundancy on multiple servers in geographically diverse locations, however am not particularly familiar with what hardware / software and other things would be involved.

You can get smart load balancing devices (re: other thread) that will handle routing the traffic geographically I believe, the difficult bit is keeping the data in sync across servers.

othellotech · 26th August 2003

the simple answer is that it's not easy and it's not cheap.
the long convoluted answer is that it's going to be expensive and take a lot of planning

it depend son if you want high-availability or load balancing

HA is *realtively* straightforward, hot standby machine, rsync the contents, internal and external monitorring systems, and enginerer to hand to bring it up on a differnet IP when the primary fails

load balancing requires some hardware, and in practice, to do across multiple locations you need to look @ loadbalancing clusters of customer facing servers which read the site(s) off a variety of back-end servers at multiple locations - it depends what you're trying to acheive and how many pages left in your cheque book :P

M@tt · 26th August 2003

The customer that wants it is prepared to pay, I am told.

They have a MySQL driven site which adds to the fun factor.

For a small budget (relatively speaking, as it will be expensive), what would we be looking at paying and what kind of hardware would we need?

They want the site mirrored from a server in the UK to one in the US, and if either one goes down for the other one to take over the hosting of their website. When both are up, they balance the load.

racksense · 26th August 2003

Quote:

Originally Posted by M@tt

They have a MySQL driven site which adds to the fun factor.

For a small budget (relatively speaking, as it will be expensive), what would we be looking at paying and what kind of hardware would we need?

MySQL v4 does replication I believe, no idea how well it works.

Something like rsync should be able to handle other data.

Look at Radware, Foundry, F5 etc for load balancing kit, I'm sure there will be others around.

Remember the simpler the design the better it'll work.

markcastle · 30th August 2003

Maybe...

Cisco Distributed Directors
Alteon Switches (Extremely hard to configure IMHO)

or... if you've got BGP4 in both locations there is a very elegant solution that could work out much cheaper than buying distributed load balancing switches. Do a search for "bgp4 anycasting"

If it wasn't dynamic content i'd have said look at Akamai

racksense · 6th September 2003

I forgot about, but was reminded of elsewhere, http://www.wackamole.org/

Looks very good, especially with Backhand.

othellotech · 6th September 2003

the prolems with wacamole are that
it doesnt replicate content - so you have to implement a replication methood across all the servers
it doesnt really balance load - relying on round-robin
all the machine need multiple ip's ( the *real* ip it communicates with and the virtual ip the website www A record is assigned.

its good in that it ensures that theres alway a machine responding on every ip but its not really redundancy ...

racksense · 6th September 2003

The data replication is a separate issue, you're going to have to have some form of replication whatever technique you use for the actual balancing.

I think it can provide load balancing within each cluster, for some applications, eg. mod_backhand.

Usage of multiple IPs is going to happen one way or the other (real or RFC1918) which ever solution you use.

How is it not redundant when it ensures that a particular IP is always providing service?

I think it's a pretty cool solution for some applications, have say one cluster in the UK, one in the US, with DNS pointing to both clusters. Anyway, it all depends on exactly what you're trying to achieve, "load balancing" is a very wide ranging function..