With HTTP caching frox rewrites ftp retrieve requests in HTTP and sends them to a proxy server like squid which actually retrieves the file and does the caching. With local caching frox maintains its own cache of recently downloaded files on the hard disk.
The main advantage of the HTTP method is that the cache can be shared with any web browsers which use the same proxy directly for their ftp requests. Local caching is much faster (especially for small files) and lighter on network resources.
Caching needs to be compiled in by running
--enable-local-cache. You can have both compiled in if
you want - the cache method actually used (if any) is chosen
in the config file.
For HTTP caching your config file needs to say:
CacheModule HTTP HTTPProxy 192.168.2.1:3128 # proxy address # and optionally... MinCacheSize 64000 # in bytes
Files smaller than MinCacheSize will not be cached. Latency on small files can be really bad with this method since the HTTP proxy has to make a separate connection to the ftp server - setting MinCacheSize fairly high will help with this.
For Local caching your config file should say:
CacheModule Local # and optionally... CacheSize 100 # in MB
The cached copies will be created within frox's working directory as specified in the config file. If CacheSize is unspecified the cache will grow indefinitely.
Finally you can turn off caching with the line
None. This is necessary if you wish a config file subsection to
not do caching when the parent section has it turned on.
With strict caching off frox works out the URI for the file it is retrieving by sending a PWD to ask the server what the current working directory is. On a small minority of servers this may be broken, and a server may return the same answer to PWD for more than one directory. If additionally two such directories contain files with the same names, sizes, and times of modification then frox could cache the wrong file by mistake.
StrictCaching tries to get around this by keeping track of all directory changes and making up its own path instead of asking the server for it. This has the disadvantage that ``cd pub; get readme'' will not cache to the same file as ``get pub/readme'' or ``get /pub/readme'' or ``cd /pub; get readme''. As a result your cache hit rate will probably drop considerably, but you will have the extra security. Note that you can set this on a per server basis if there is a particular server about which you are worried.
Thanks to Henrik Nordstrom for pointing this problem out and suggesting a solution.
This option involves using FQDNs instead of IP addresses for the cache requests (either to squid or the local cache), and allowing squid to guess whether to use binary or ascii file transfer rather than using the value supplied by the ftp client. For transparently proxied connections this requires a reverse DNS lookup to get the FQDN, and may cause problems with un-synced mirrored ftp hosts all running off the same FQDN, although most of the time there will be no problems. The advantage is that you should increase the hit rate, in particular with squid based caching for files which have been previously downloaded through squid with a web browser.
If all your clients are being non transparently proxied then this option should cause no problems. Otherwise use with some care, although it may be worth the extra hit rate.
Check that frox was compiled with support for the cache module you are attempting to use, and that you have given the correct Cache options for that module. If doing http caching check that the http proxy name resolves.
Files are not cached if the ftp server doesn't support the
MDTM and SIZE extensions. Non-Anonymous ftp sessions are only
cached if you have set the
CacheAll config file
option. Additionally if the host's DNS name resolves to more
than one IP address then the file will be cached, but frox has
no way of knowing that the different IP addresses refer to the
same host - unless your ftp client happens to pick the same IP
both times you will not get the cached file back. Using
CacheOnFQDN may impove this behaviour.