Alex Rousskov writes:
>On Wed, 28 Oct 1998, Niilo Neuvo wrote:
>
>> The other point is that I want to move the process of selecting the
>> cache that actually contains (or should contain) the object to the
>> client software.
>>
>> I have a small javascript code that does this by calculating a hash
>> code from every url and decides what parent to use. It is vital that
>> the javascript uses the same algorithm as the parent cache uses.
>
>Hmm.. Change your custom Javascript hash function to mimic CARP hashing
>algorithm and use CARP on parent? CARP support in Squid is experimental, but
>I think people reported that it works. Just a thought...
The capabilities of javascript are very limited, especially on the
older browsers. For example I can't convert characters to integers,
which is vital to most hashing algorithms (like the one CARP uses).
Nor can I perform floating point math and even dream that the machines
will get same results (this is a very small problem).
Otherwise CARP seems to almost do what I want. I'll just write my own
version of carp.c.
Having similar hooks as Apache to various parts of the engine would be
nice. Even if it were implemented with dlopen() as I was planning to
do.
After reading through carp.c I encountered the following unrelated
problems:
/*
* sum of carp-load-factor's for all cache_peer's in squid.conf
* must equal 1.0
*/
assert(a == 1.0);
This floating point math will never work out if I have three caches
that should be equally balanced. In my opinion math like this should
be written with integers.
Line 93 in carp.c calls ulrCanonical, which returns a malloced string.
This is never freed. I haven't noticed anything about a garbage
collector, so this probably is a bug.
-- Niilo Neuvo >>>^<<< niilo.neuvo@saunalahti.fi Technical Director /¤\ +358 (0) 50 5611042 (mobile) SAUNALAHDEN SERVERI OY | | +358 (0) 3 31558042 (fax) .¸¸.·´´¯·.¸¸.·´´¯·.¸¸.·´´¯·.¸¸.oOOOo.¸¸.·´´¯·.¸¸.·´´¯·.¸¸.·´´¯·.¸¸.·´Received on Tue Jul 29 2003 - 13:15:54 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:57 MST