On 02/13/2014 03:58 AM, Dr.x wrote: > hi all , > im asking about memeory caching > now we know that we have restirction with smp rock to 32 k , whiuch result > a poor bw utilization It is really not clear what you mean by "poor bw utilization" and how it is related to 32KB object size limits in small rock. I recommend that you use a different terminology or explain the connection between the two concepts so that it is clear what you are asking about. Perhaps you mean poor byte hit ratio? > now > > what about memory object size ? > is is also 332 k ? Shared memory cache prior to Large Rock support has the 32KB limit. Shared memory cache after Large Rock support does not have that limit. Non-shared memory cache does not have that limit. > i mean if i get a server and depened on ram and have 128G ram , and dont > want to use hardsisks > > does in this case we still have 32 k object size max in ram ?? Depends on the memory cache type. See above and shared_memory_cache in squid.conf.documented. > what is the best implementation for smp with rock with aufs so that we have > max bw utilization and max hit ratio ? Rock and aufs are two different cache store modules. You cannot use "rock with aufs". Rock was designed for SMP use. The current aufs store implementation is not really compatible with SMP. The best store to use depends on many factors. For example, you can use Large Rock store to avoid the 32KB limit but you will have to run Squid trunk, with all the caveats that implies. HTH, Alex.
