Re: Summary: Preventing Transfer Delay to DR-11

From: Guy Harris (guy@Auspex.COM)
Date: Thu May 07 1992 - 16:25:49 CDT


>It's worth noting that SunOS 4.0.3 does not have Shared Objects/Libs,
>thus despite the fact that we had only 8MB of RAM, we had no problems
>with swapping. I claimed 5 buffers of 128MB of shared memory for this
>process.

It's worth noting that he presumably means that 4.0.3 *DOES* have shared
objects/libs, because it definitely does....

>If you run set-uid-to-root, you can do more, such as:

...

> - use asynchronous I/O.

There doesn't appear to be any code in "aioread()" or "aiowrite()" or
anything it calls that *requires* super-user privileges to do
asynchronous I/O, so it's not clear why it requires that the program run
set-UID to "root", if "asynchronous I/O" means "aioread()",
"aiowrite()", and company. In fact, it *doesn't* require super-user
privileges....

You *do*, of course, have to configure async I/O into the system, if
that's what he meant by "asynchronous I/O".

> This is the biggest potential win, the
> driver has to allow setting up the pending DMA and switching
> to it at interrupt time. That would shorten latency to ~30 microsec.

With the "aioread()"-and-company async I/O, the driver hasn't any idea
that async I/O is being done; it's done by handing the I/O request to a
kernel thread which does the request as an ordinary synchronous I/O
operation.

Now, if you're referring to modifying the driver to do async I/O itself,
yes, that'd probably require root privileges, so that you can reboot the
machine with the kernel with the new driver, or load the new driver with
"modload"....



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:42 CDT