Network Booting versus hard disks: Costs and Implications

In the African context, it does not necessarily make sense to spend large amounts of money on proprietary computing solutions when low cost solutions exist. This paper explores the experiences we have had in our University implementing both network-booting and hard disk- booting Linux systems. Most Linux systems boot from hard disk. Our system, one of the few in the world, boots mainly from ethernet, completely without hard disks. We are not aware of any such system of this scale; we have approximately 200 computers using this system. We discuss cost implications, maintenance issues, usability issues, and performance issues. We have found that in terms of both hardware and software costs, this is the cheapest possible solution. In terms of maintenance, the initial investment is high, but once the system is established, it is extremely quick to add another client computer. In terms of usability, the variety of Open Source software now available renders the system highly usable for both our specialised needs as well as day-to-day computing. The primary difficulty we experienced was around performance. We found that performance statistics varied greatly and were highly susceptible to the number of users on the system, the number of servers, how the server load was distributed, the software networking configuration, and the protocols used. In particular, we found that dividing the server load up, carefully, had the single biggest impact on the system performance. Another factor which severely impacted the performance was overall network load as determined by the protocols and network software configuration. A well-configured client computer in a lightly-loaded server environment could boot in about 30 seconds. A poorly-configured client in a busy environment could take half an hour. Hard disk boot time performance is typically of the order of slightly under one minute. Thus, in optimal circumstances, network booting is faster than hard disk, whereas in sub-optimal circumstances the performance is worse. We believe that the maintenance overhead and cost savings make the solution viable, but, for smaller sites, we recommend against our solution and advocate a hard disk based solution.