Weird I/O slowdowns after kernel upgrade

I recently updated my kernel from 2.4.19-4a to 2.6.11.11 (big jump, I know) and so far haven't had any problems except for one rather major one... any operation dealing with I/O (screen, network, disk, even loopback devices) seems to take much longer to complete than usual, and meanwhile the entire system loses responsiveness almost to the point of unusability. I'm on a PowerPC, if it matters.
I'd post more information, but I don't know what would be relevant and what would be useless.
-:sigma.SB

P.S. It's almost impossible to use Mozilla now. I ghosttyped this entire post. As I write, it's gotten to "much longer to complete."

If it were a PC I'd say that you have problems with DMA,
but on a PPC ... *shrug* ... sorry, no clue. I'd go back to
the old kernel, and start over.

How did you compile the new kernel, copy 2.4s .config,
make oldconfig?


Cheers,
Tink

I haven't got access to my old kernel's configuration, so I had to reconfigure the kernel from the bottom up.
It may also be relevant that I recently upgraded this machine from 64MB to 320MB using a RAM upgrade from a dead PowerBook, which I'm not sure is "supposed" to be compatible with my machine. Nor am I entirely sure that I installed it properly, for that matter. Mac OS 9 also exhibits similar unresponsiveness when running memory hogs (such as Internet Exploder, though at the time I had assumed it was just because it sucked) and I wasn't running 2.4 long enough after the upgrade to make sure it was running perfectly. Since most of my I/O hungry programs are memory hungry as well, this may be the actual problem.
As I cannot afford a new RAM upgrade, if this turns out to be the case, the question I must ask myself is... am I better off with swap, or weirdly latent RAM?
An alternate answer might be if the length of a quantum was reduced between 2.4.19-4a and 2.6.11.11, which might cause problems due to my slow processor; is this a configurable (or hackable) option?
-:sigma.SB

Originally posted by Solra Bizna I haven't got access to my old kernel's configuration, so I had to reconfigure the kernel from the bottom up. It may also be relevant that I recently upgraded this machine from 64MB to 320MB using a RAM upgrade from a dead PowerBook, which I'm not sure is "supposed" to be compatible with my machine. Nor am I entirely sure that I installed it properly, for that matter.

couple of thoughts;
- first up try running memtest86 for a couple of hours
- see if you can extract the config

zcat /proc/config.gz > .config
make oldconfig

- first up try running memtest86 for a couple of hours

Would memtest86 work on my PowerPC?

- see if you can extract the config

I tried that, no luck.
-:sigma.SB

Originally posted by Solra Bizna Would memtest86 work on my PowerPC?

Oooops - missed that...

The config extraction was a stab on the dark - handy if it's there.

[ot]I'm posting this from Lynx. Mozilla is simply unusable.[/ot]
I hacked together a quick benchmark program which gives itself realtime priority, and then runs a benchmark on memory blocks that increase in size. The operation it performs is to read a (32-bit, unsigned) value from memory, increment it, and write it to the next 32-bit slot in the block of memory. It continues until it has run through the entire block 10 times.
This is the output of this benchmark (unmodified) on blocks from 1KB to 256MB on kernel 2.4 vs. 2.6.
http://sigma.tejat.net/membench.2.4.txt
http://sigma.tejat.net/membench.2.6.txt
Under 2.4, I get about 100 nanoseconds of response time per byte, but on 2.6 I get about 225. Something is wrong here.
-:sigma.SB