Re: [WAS] Re: upgrading to v2

On Thu, 8 Jul 1999, Hossam El-Ashkar wrote:

]> Without looking in too much detail, I'd reduce cache_mem to 8 MB (the
]> default). Linux is going to buffer all your ext2 data anyway. There's
]> not much profit in doing it twice
]
] Is this applicable to Solaris too?? Or just for Linux on ext2 file
]system?? I am having a 1GB RAM machine. Won't increasing cache_mem be
]profitable??

It would apply to almost any kind of OS which supports the kernel buffer
cache concept. Well, it applies to Solaris, too, though it offers the
forcedirectio mount option. From the mount_ufs man page:

 [...] If the filesystem is mounted using forcedirectio, then data is
 transferred directly between user address space and the disk. If the
 filesystem is mounted using noforcedirectio, then data is buffered in
 kernel address space when data is transferred between user address space
 and the disk. forcedirectio is a performance option that benefits only
 from large sequential data transfers. The default behavior is
 noforcedirectio.

Hence, it would seem that you should increase your cache_mem, if using
forcedirectio. But caches don't transfer large sequential data! And some
measurement I did back in the squid-1.2 days did not indicate anything
above measurement noise from using forcedirectio, though. Thus I'd suggest
going by the squid gurus' recommendations.

Le deagh dhùrachd,
Dipl.-Ing. Jens-S. Vöckler (voeckler@rvs.uni-hannover.de)
Institute for Computer Networks and Distributed Systems
University of Hanover, Germany; +49 511 762 4726
Received on Thu Jul 08 1999 - 11:17:07 MDT