Re: Sequential sever?

On Oct 2, 4:49pm, James R Grinter wrote:
> Subject: Re: Sequential sever?
> Alex Rousskov <rousskov@ircache.net> writes:
> > Except that all _disk_ operations/handlers are blocking. Select/poll(2)
> > always return "ready" on a disk file descriptor. Blocking disk I/Os were
> Absolutely (and yet someone spent a real long time writing the
> original code as if they were able to be non-blocking.)
>> the motivation behind introducing real threads into Squid (the now
>> officially dead async-io stuff).
> oh, this is a shame - I've obviously not been following closly enough,
> recently. Is anyone looking at implementing in a different manner (eg
> multiple processes to schedule the reads/writes in a way akin to the
> dnsservers)?

It may be a crazy thought, but has anyone considered using a "raw device"
mechanism for the cache. It could posibly be async, and might be able to
avoid the file descriptor and fragmentation problem. Informix IDS works this
way and seems to gain considerable performance by avoiding the file system. A
raw device is simply a hard drive parition, or several harddrive partitions
that are opened by the program and used directly instead of via a filesystem.
Received on Sat Oct 02 1999 - 10:43:47 MDT