Hi, I recently discovered CARP[1] which is implemented in Squid[2] and provides the same features as straw2: - stable placement when a device is added or removed - weighted devices - given enough objects their distribution is uniform The differences are: - the hash function: - bit rotation based on the name of the item / object for CARP[3] - Robert J. Jenkins hash on the PG id for CRUSH[4] and no hashing for the item id - the combination of the item hash and the object hash - XOR for CARP[5] - Robert J. Jenkins hash on the PG id hash and the item id for CRUSH[6] - applying the weight to the the combination of the item hash and the object hash - the multiplier is calculated per device for CARP[7] and the combined hash is multiplied by it to get the score[8] - the weight is multiplied with the log of the combined hash for CRUSH[9] I'd like to setup a simulation that shows the effect of these differences. I tried a few but the results are roughly the same wether I use CRUSH or CARP. Any idea ? Cheers [1] CARP https://tools.ietf.org/html/draft-vinod-carp-v1-03 [2] CARP in squid http://bazaar.launchpad.net/~squid/squid/trunk/view/head:/src/carp.cc#L198 [3] CARP hash http://bazaar.launchpad.net/~squid/squid/trunk/view/head:/src/carp.cc#L96 [4] CRUSH hash https://github.com/ceph/ceph/blob/master/src/osd/osd_types.cc#L1357 [5] CARP combination http://bazaar.launchpad.net/~squid/squid/trunk/view/head:/src/carp.cc#L200 [6] CRUSH combination https://github.com/ceph/ceph/blob/master/src/crush/mapper.c#L313 [7] CARP multiplier http://bazaar.launchpad.net/~squid/squid/trunk/view/head:/src/carp.cc#L133 [8] CARP score http://bazaar.launchpad.net/~squid/squid/trunk/view/head:/src/carp.cc#L203 [9] CRUSH score https://github.com/ceph/ceph/blob/master/src/crush/mapper.c#L335 -- Loïc Dachary, Artisan Logiciel Libre -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
