Re: cache sizes & needed ram.

On Jul 16, 9:36pm, Dave Zarzycki wrote:
} Subject: Re: cache sizes & needed ram.
}
} Keep in mind that you still need RAM to cache objects and for objects in
} transit! It would be wise to allocate an equal amount or more just to
} cache files. Now add on the RAM necessary to keep your OS happy...

I think it would help immensely if squid didn't keep large in-transit
objects in memory. I don't see any reason that it couldn't flush the
part of the object that has already been transmitted to the client from
memory. If the data is arriving faster than the client is reading it
and memory is tight, squid could even flush the data between the read
pointer and the write pointer from memory and burn an extra fd to read
the data back from disk to send it to the client while it is appending
to the disk file. This would allow much better control over the
memory footprint by using the cache_mem parameter and avoid large amounts
of memory getting consumed by in-transit objects.

I don't want to go all the way to the NOVM version, since it has to
hit the disk for small, hot objects. I don't think that would make
a very good web accelerator, since the web server would be doing
pretty much the same thing. The open() system call is the bottleneck.

                        --- Truck
Received on Thu Jul 17 1997 - 01:28:24 MDT