Competitive advantagesHardware clustering versus software clustering
Advantages and disadvantages
Let us examine the two existing types of hardware clustering to execute the load balancing and data high availability brought in by the SafeKit software clustering solution.
S |
Network boxes (load balancing) | Cluster hardware (data availability) | Software (both) |
 |
 |
 |
|
| Target | Web farms | Databases | Small/medium applications |
| Best for | Large web portals with load balancing | Large database on shared disk | Easy deployment with standard servers |
Server farm with network load balancing and high availability
With a hardware clustering solution, network load balancing is implemented using specialized boxes. Two redundant network boxes are added upstream a server farm. They distribute network traffic to the server farm and isolate a faulty server.
With the SafeKit software clustering solution in a farm architecture, no network hardware is required to implement load balancing. Load balancing is provided very effectively on each application server by filtering inbound packets (cost < 3% CPU), without any need to add load balancing hardware.
In terms of price and functions, the farm solution offered by SafeKit is very suited to small or medium-sized farms, typically with 2 to 10 servers. Beyond that, for very large farms, investing in costlier network-load-balancing boxes becomes justifiable.
Hardware cluster with high data availability
With a hardware clustering solution, data high availability is ensured by a disk bay placed between two servers. The data of a cluster application is located on a shared disk in the bay and, in case of failure, the disk is switched from one server to the other, and the application restarted. This type of solution is complicated to use and maintain because of its strong reliance on the disk bay for switchovers.
With the SafeKit mirror software clustering solution, the disk bay is not required between both servers. Instead, the SafeKit software is installed on the two servers and replicates data through the network between the internal disks of the first server and second server.
It is configured without any disk-organization constraints on both servers: you only need to define the file directories to replicate. SafeKit only replicates modifications made in files opened by a running application, thus limiting traffic. The synchronous data replication ensures perfect mirror files between the first and second servers’ internal disks.
This way, in case of failure, the application starts on the second server from the last data backed up on the disk and, thus, without data loss unlike an asynchronous replication solution.
The SafeKit mirror solution with file replication is very suited to small or medium-sized databases, typically with less than 100 Gigabytes and less than 200,000 files. For larger databases, hardware clustering solutions with shared disk bays and internal replication systems are more appropriate.
Nevertheless, SafeKit adapts very easily to disk bays shared in NAS mode by retaining a very simple solution and thus exceeding the limits on size and number of files.
More information:
