Ye Old Laptop as a Server
Resourceful people can use an old desktop as a server of some sort, but they also may have an old laptop around collecting dust. Although most of today's desktop environment software runs too slowly on these older laptops, you may not want to get rid of yours yet. These cast-offs easily can be transformed not only into a low-power server, but also into a mobile VoIP and Web server.
Let's look a three Linux distros with this in mind and test two different Web servers. The Ye Old Laptop used in this article is a Dell Latitude with a 1.0GHz Intel Pentium III Mobile CPU and 256MB of RAM. Keep in mind that Web 2.0 software may be too much for these old cast-offs. So, if you need to serve the latest-and-greatest flashy content, you're better off buying a new desktop computer. The server configuration I used and tested on every Linux distribution is Apache 1.3.34, Lighttpd 1.4.13, PHP 5 and MySQL Server 5.
Xubuntu is an official Ubuntu distribution for old or slow systems. The main difference between it and Ubuntu is that Xubuntu uses the Xfce window manager instead of GNOME. An X-less Ubuntu or Kubuntu installation should have the same performance results as an X-less Xubuntu. With X running, Xubuntu requires at least 128MB of RAM, but the minimum requirements for the Xubuntu Alternate CD without X are 64MB of RAM. It uses the Ubuntu repositories, which are filled with programs that satisfy nearly every system administrator's needs. I installed Apache 1.3 and Lighttpd 1.4, MySQL Server 5 and PHP 5 with apt-get install apache lighttpd and apt-get install mysql-server-5.0 php5. For some reason, I had to install Apache 2.0 (and all its dependencies) to get PHP to work with Apache 1.3. After I got it working, I then uninstalled Apache 2.0.
After configuring Apache and Lighttpd, I stress-tested the Web server over a 100Mb Ethernet connection with Siege, a benchmarking and HTTP regression-testing utility. It is designed to let Web developers measure the performance of both their code and their servers to see how well the server functions on the Internet under a heavy load. Siege supports basic authentication, the HTTP and HTTPS protocols, and even cookies. This utility allows system administrators to hit their Web server with both a concurrent and configurable number of simulated users.
I generated a 5KB file with dd and ran a stress test with siege 192.168.1.179/5kb -c10 -r10 on a fairly new computer, simulating ten clients refreshing ten times each second (a fairly heavy load). To improve results, I turned off all unneeded services, including Xorg, Samba, the FTP server and the IRC server. If you run Siege with the -b command-line argument, it runs the test without any delay for throughput benchmarking. This means that the server gets all the hits at once (but that isn't a very realistic scenario). Without the -b flag, Siege invokes each simulated user with a delay of at least one second. Table 1 shows the performance differences between Apache 1.3 and Lighttpd 1.4.13.
Table 1. Performance Differences between Apache 1.3 and Lighttpd 1.4.13 on Xubuntu with a 5KB File
| Ten clients, ten hits/second, 5KB file | Apache 1.3 | Lighttpd 1.4.13 |
|---|---|---|
| Time elapsed | 9 secs | 8 secs |
| Data transferred | 0.2MB | 0.35MB |
| Server response time | 0.06 secs | 0.00 sec |
| Transaction rate | 11.00 trans/sec | 12.5 trans/sec |
| Throughput | 0.02MB/sec | 0.04MB/sec |
| Concurrency | 0.7 | 0.02 |
| Longest transaction | 3.00 secs | 0.04 secs |
| Shortest transaction | 0.00 secs | 0.00 secs |
The tests performed on Xubuntu 7.04 (Table 1) reveal that the transaction rate is high, and the shortest transfer time is about a millisecond or less. (Siege can't calculate anything below a millisecond.) This is mainly because my network consists of only 20 feet of cable. Depending on your connection, testing over the Internet will show a greater latency than over a local network. According to the Siege man page, concurrency is the average number of simultaneous connections. When the number rises, the server performance decreases, especially on slower hardware.
Table 2. Performance Differences between Apache 1.3 and Lighttpd 1.4.13 on Xubuntu with a 1MB File
| Ten clients, ten hits/second, 1MB file | Apache 1.3 | Lighttpd 1.4.13 |
|---|---|---|
| Time elapsed | 12 secs | 12 secs |
| Data transferred | 108.41MB | 108.40MB |
| Server response time | 0.5 secs | 0.4 secs |
| Transaction rate | 8.3 trans/sec | 8.2 trans/sec |
| Throughput | 9.0MB/sec | 8.9MB/sec |
| Concurrency | 4.3 | 3.5 |
| Longest transaction | 1.30 secs | 0.75 secs |
| Shortest transaction | 0.1 secs | 0.1 secs |
The tests performed on the 1MB file (Table 2) show that under a heavy load, Lighttpd 1.4.13 performs the best with smaller files (such as blog posts without any pictures). Performance, however, is still higher than Apache with 5KB files. Interestingly, the heavier the load, the better Apache performed. Results from tests on 5MB files under a very heavy load of 50 clients at ten connections per second showed a 0.1 second higher response time and a 0.1MB/sec higher throughput than Lighttpd. Lighttpd still had shorter transaction times that were four seconds less than Apache though.
On Xubuntu, Lighttpd is the fastest performing server for static documents, although Apache follows close behind. For best results in Xubuntu, turn off all unneeded services, so that if the server gets a heavy load, it will have a few more resources to improve speed. I found Xubuntu to be quite fast and easy to get running, but there are other distros to consider, such as Vector Linux.
Trending Topics
| You Need A Budget | Feb 10, 2012 |
| The Linux powered LAN Gaming House | Feb 08, 2012 |
| Creating a vDSO: the Colonel's Other Chicken | Feb 06, 2012 |
| Your CMS Is Not Your Web Site | Feb 01, 2012 |
| Casper, the Friendly (and Persistent) Ghost | Jan 31, 2012 |
| Razor-qt 0.4 - Qt based Desktop Environment | Jan 30, 2012 |
- 100% disappointed with the decision to go all digital.
- Linux--The Internet Appliance?
- Kernel Korner - The New Work Queue Interface in the 2.6 Kernel
- Kernel Korner - I/O Schedulers
- UpFront
- Readers' Choice Awards 2011
- Why Python?
- Real-Time Linux Kernel Scheduler
- Solaris-Zones: Linux IT Marbles Get a New Bag
- You Need A Budget
- Search
3 hours 35 min ago - Question
3 hours 58 min ago - for the record
4 hours 48 sec ago - That's disappointing. Thanks
6 hours 24 min ago - Well spotted. I've corrected
7 hours 53 min ago - This is a great program. We
10 hours 53 min ago - No Air for Linux
12 hours 42 min ago - HEWLETT PACKARD created
12 hours 52 min ago - HEWLETT PACKARD created
12 hours 55 min ago - very helpful :)
13 hours 17 min ago






Comments
Ye old laptop server with OpenBSD
Trying a similar tactic, but this time with OpenBSD on an older Dell Latitude with less memory. andyviar.homeip.net