Article:
 |
|
Why Install Linux on Your Mac?
|
| Subject: |
|
speed - But Macs Are Slower, Right? |
| Date: |
|
2004-12-01 12:05:54 |
| From: |
|
williamverna
|
Response to: speed
|
Quote by Paul Murphy, LinuxInsider
09/30/04
"The Mac runs Unix on RISC. That makes it inherently more secure and more reliable, as well as more attractive, to people whose jobs or interests require them to go beyond the limits of the GUI.
To many people in the sciences and engineering... the PC versus Mac speed issue boils down to hardware running Unix -- an arena in which UNIX BSD on the G5 has a 50 percent advantage over Linux on the (Intel Processor) Xeon."
Simply put BSD Unix on a PPC is faster and more stable then Linux on x86. Darwin has extremely fast I/O and IP stack, so I have serious reservations believing that a Linux kernel and distrobution ported to a PPC would be faster then the FreeBSD Unix subsystem and Mach kernel used in Darwin.
http://www.linuxinsider.com/story/36964.html
|
Showing messages 1 through 1 of 1.
-
speed - But Macs Are Slower, Right?
2004-12-02 01:28:16
wapentake
[Reply | View]
Your serious reservations are naive. As I said, benchmark it yourself. Or go read the conference proceedings from SOSP in the 90's, and see the papers rip Mach apart, and show why Mach gave microkernels a bad reputation. Microkernels can be fast. Mach is the opposite of fast. Just look at the source code of Mach and BSD in Darwin: they are not designed for speed, but for software management (i.e., to keep things well modularlized). Just one example: in Linux, the parameters for a system call are passed directly via the register file. In Darwin, they are packed up in a memory buffer, passed around, and the results are passed back via another memory buffer, and then everything written back to the register file. That just does wonders for the cache and TLB.
By the way, the I/O and IP stack on Darwin is not fast. It is easy to verify by benchmarking; you can convince yourself rather than believe me. And Darwin doesn't even scale beyond a single processor. When you add Apple's second processor (ever wonder why they don't sell more than two processors?), the performance gap between Linux and Darwin will probably become larger (I haven't measured this myself), but it is easily evident, since Darwin uses several big-kernel-locks, while Linux uses fine grained locking. If the IP stack of Darwin runs on one CPU, it can't for the most part run concurrently on another CPU. Likewise for the I/O system. Fast, huh?
I have an idea for an alternative demonstration of Darwin's diminutive performance. There is currently a project to add Darwin compatibility to BSD and Linux. Go find this project, and try it out. You'll see, without a doubt, that Linux is still faster than Darwin when running Darwin applications, all on PowerPC.