Hardware
Wed, 2010/10/13 - 11:54
- Servers -- these can range in prices but a reasonable metric for a fully redundant, hot-swappable component-based server is about $5,000 per CPU and 4GB of RAM . You might get there with blade technology, or standalone rack-mounted servers, or as a piece of a much larger, multi-processor beast. All have advantages and disadvantages that depend mostly on the in-house IT skills
- Load Balancers -- any web-based service eventually needs a load balancer for scale and redundancy. You have two choices that I once labeled incorrectly as software and hardware when what I should say is dedicated versus distributed. Dedicated (or hardware) load balancers are the preferred approach in a large installation, not just for performance but also for administrative benefits. Distributed (or software) load balancers run on all servers involved in the cluster and each is aware of and participates in the cluster; they all must run the cluster/load balancer software which may interfere with the ECM application
- Firewalls -- a necessary evil, a firewall or two must be in place to restrict as much unwanted, unneeded network traffic as possible. Edge firewalls should be dedicated devices, each server may also have local firewalls to restrict access to ports (both in and out). Single applications can easily use one of the many free firewalls available from many Linux distributions, software from firewall vendors is very expensive and complex (and usually necessary in any environment seriously looking at ECM). They are usually easy to configure and understand and will offer automatic updates (not necessarily a good idea for a firewall of course). You really want two firewalls that offer at least fail-over services if not load-balanced services. There are many good, relatively cheap, dedicated firewalls available that meet these requirements.
- Router -- Possibly not required or as a part of the firewall, something performing network routing is required to connect your network to the Internet at large. This needs to be redundant or when it fails all services are out. Routers are not necessarily expensive devices, however dynamic routing like BGP is an advanced IT task and is usually done on very advanced routing hardware.
- Switches -- the servers on your network need to communicate with each other and the load-balancers/firewalls. You'll need at least two physical devices to achieve redundancy, even with the use of VLANs.
- hard drives -- Storage is one of the most variable costs involved in any computer-based application, but even more so for Enterprise Content Management because "Content" comes in all sorts of different types and sizes. You must ALWAYS be concerned about data integrity, you need to look hard at redundancy and fault-tolerance, there are many solutions available that provide solid data integrity using relatively cheap, commodity drives. However, that only works well with low-traffic / low-volume sites. Performance is really the metric to be concerned about. Dont be fooled, though, unless you are Facebook-sized, NAS is your friend and iSCSI is the next obvious place to look. Fibre Channel is dead technology, if you still have some then go ahead and use it, but iSCSI is a far more cost-effective protocol.
