Hi, I'm running a 266 MHz K6 linux box with Debian 3.1. I've been getting slow performance running programs like Firefox or multitasking and am looking into cheaply upgrading it. My hard drive is obviously very slow, but I'm not sure my motherboard/chipset can handle a modern UDMA disk, so I was considering memory. I came across this article: A Primer on DDR Memory, which talks about memory use with various versions of Windows. It talks about cost vs. benefit for adding RAM, citing a drop in performance if too much RAM is used. It says 64 MB is about the peak for performance with Windows 95 (the former OS of this box).

My question is whether the same applies for Linux. If I add 128 or 256 MB of RAM, will I see an improvement over my current 32 MB or will it start degrading performance after 64 MB or so?

Thanks, anybody who can clear this up for me. I'm hoping this is just a flaw with Windowa because I've heard Linux is good about using whatever RAM it can.

definitly add ram
no perfomance drop
older 2.4 kernels don't go over 1GB too well but 2.6 can use PAE extensions well and can use 64GB

oh yea Windows is a toy not a computer so it's designed to run toy size ram

It's also possible that the hard-disk access parameters are set very conservative, hence very slow. Try man hdparm and also look for hdparm as a keyword both here and in Google.

Memory is pretty much a no-brainer purchase as far as performance goes. Get Some. More.

Basic rule of thumb: If it doesn't work well in Windows its because Windows sucks rather than being a general limitation of all Operating Systems.

Okay, thanks, all. I'll get me some, heh.

I'm also trying the Opera browser and it's loading about ten times faster. Too bad menus look like crap and there's ads. Hopefully Firefox will run better with more memory.

I'll also look into my harddrive parameters as well as my actual hardware limitations. If I can use a faster drive, I'll do that.

Thanks again!

EDIT:
hdparm indicates that multcount is set to 16 and it's using DMA. It's benchmarks were
"Timing cached reads 136 MB in 2.05 seconds = 66.34 MB/sec
Timing buffered disk reads 18 MB in 3.25 seconds = 5.54 MB/sec"
I tried changing a couple settings but the benchmarks suffered (cached: 4.53 MB/sec and buffered: 4.27 MB/sec). Looks like it's running about optimized as it is. Oh well.

Good idea. When computer manufacturers design retail-grade computers, they know that those machines won't be doing much I/O. So they often design very low-grade IDE channels and put them "conveniently" on the motherboard. And they sell disk drives that, while extremely capacious, are also very slow. The machine works, but it can barely get out of its own way once the signals have to leave the motherboard. (And sometimes there is a rather shocking difference between how fast the CPU is said to perform and how fast the board can perform... and of course "guess who wins.") The biggest bottleneck is, and always will be, I/O. (By adding RAM you are basically "avoiding I/O.")

You might be able to transform the machine outright by:

1. Adding more RAM.
2. Installing a separate EIDE-controller card.
3. Installing a USB-2.0 / Firewire card.

The folks who design off-board peripheral interfaces expect that many of the people who buy them "want more." Now, IDE (no matter how you slice it) just isn't tremendously-fast just because of its design. But if you want to watch a piece of hardware jump, try using Firewire. This is the latest incarnation of SCSI and ka-a-a-a-zing!