2005 Why Do Clusters Suck?
From Cluster Documentation Project
The following were comments received in our 2005 Survey (11 respondents, reponses are grouped by respondent)
- systemic component failures
- enough clout to influence the component vendors
- Disk less computing with oneSIS
- Cluster management has to be one of the areas where many clusters still really suck
- Bios, bios, bios!
- Commercial License issue
- High-Speed Interconnects are the disabling technology of clusters.
- The learning curve for end users is more like a learning mountain.
- Applications on clusters tend to run at 1% to 5% of "peak"
- Clusters are a systems management nightmare because they are assembled from lots of parts and no one is responsible for all the parts as a coherent system.
- Status and Message Logging in Clusters
- Booting Requirements and Principles
- Clusters as Unified Systems
- because of the ill-fitting components that they are constructed with. While technically capable, you wouldn't haul iron-ore with a bunch of Ford Festivas. There are a number of things that manufacturers could make standard and make clusters suck less
- Another reason that clusters are problematic is the software that you use on them is often hard to build and configure for the cluster.
- it is very difficult to get good support staff.
- it's far from clear to me that an advance in cluster management software would result in much more productivity. As opposed to, for instance, improving network throughput, CPU power, or component reliability by a factor of 10, any one of which would lead to an immediate and dramatic productivity increase.
- My personal complaint is that there aren't enough good standard test/validation suites out there for cluster building.