February 17, 2019

Tolerating Fault in an Intolerant World - page 3

The Limitations of Clustering

  • December 23, 2002
  • By Brian Proffitt

This technology has a lot of points in its favor--but to whom is it being marketed? For that, you ask a marketing person--in this case, Mike Mitsch, Sr. Marketing Director - Server Products for NEC.

Right now, Mitsch explained, NEC is still pushing IA-64 architectures for corporate backends. Where these fault tolerant systems would come in would be in organizations that were using UNIX-based systems in the application or transactional space.

Current UNIX system users would have the least amount of hassle to jump onto an FT Linux system, Mitsch explained, because of the ease of portability of applications from UNIX to Linux.

Mitsch is quick to emphasize that application portability goes beyond that of just the operating system. As mentioned earlier in this article, not many applications can actually take advantage of clustering, thus "this is a great solution of non-cluster-aware applications," Mitsch said.

Because there is not as much of an up-front cost in switching to this new platform, there is significant costs savings. Mitsch also pointed out that there could be big cost savings in licenses (not for Linux, but the apps running on one of these FT Linux systems).

Price may be another big factor in luring people away from clusters for high-availability computing. Mitsch detailed one customer's price comparison between one of NEC's fault-tolerant systems (the 320La/Linux system for $33,693 US) versus a Sun Fire V880 for $78,990 US. Even though the customer was using Solaris applications to begin with, the costs for porting were low enough to justify the switch to the Linux platform.

Looking at NEC's approach to high-availability computing, there is a clear impression that their solution seems more elegant and focused on the essence of high availability than a traditional clustering solution, which seems to have high-availability aspects only as an afterthought.

This is not to dismiss clustering outright. For high-performance computing, clustering is one of the best and cost-efficient ways of getting the processing power needed to handle the really big problems.

But in high-availability computing, there is clearly a new alternative out there besides clustering that can be investigated by those who need it.

Most Popular LinuxPlanet Stories