-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Monday 2006-05-29 at 11:22 -0400, Bryan J. Smith wrote:
On Sun, 2006-05-28 at 20:04 +0200, Carlos E. R. wrote:
Why is that? I'm curious. Can't it use dma, even if it uses smaller writes? Just guessing.
Because all writes are Programmed I/O (PIO) through the CPU.
XOR might be simple, but attempting to deal with the traditional LOAD/EXEC/STOR through the CPU -- even using streamed XOR operations -- at hundreds of MBps is still damn slow through.
CPUs are not I/O processors (IOP). And such IOP operations take away from other operations. So unless your system is a dedicated storage device, and not servicing anything else (which is fine), you're taking away from your other operations.
And, again, it's virtually _impossible_ to measure these metrics in the current Linux kernel. All you can find out is I/O wait and servicing for processes, not the I/O latency and throughput itself.
That I understand. But my question was in the line of why raid-5 uses, or has to use, PIO instead of DMA. What is so different for raid-5. I can guess, but I don't know. - -- Cheers, Carlos Robinson -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFEe0BGtTMYHG2NR9URAmIqAJ9jq6gzzRBYVHMG3g5MtOz/DQcrlwCfTkdc zppT79H4GxKiuHek5hy6Y/I= =00IB -----END PGP SIGNATURE-----