The　JUNA　cryptosystem　is　a　new　kind　of　multivariable　public-key　cryptosystem,　which　is　evolved　from　REESSE1+.　The　security　of　a　JUNA　or　REESSE1+　plaintext　is　based　on　the　anomalous　subset　product　problem　(ASPP).　An　ASPP　can　be　transformed　to　a　special　subset　sum　problem　(SSP),　and　such　a　SSP　is　an　anomalous　subset　sum　problem　(ASSP).　An　ASSP　from　REESSE1+　is　of　low-density,　and　one　from　JUNA　is　of　high-density.　The　LLL　lattice　basis　reduction　algorithm　can　solve　low-density　SSPs　in　polynomial　time,　and　also　it　probably　solves　low-density　ASSPs.　However,　the　density　of　a　knapsack　deriving　from　a　high-density　ASSP　can　be　larger　than　1,　and　LLL　lattice　reduction　could　hardly　break　such　a　knapsack.　In　this　paper,　on　the　basis　of　our　previous　work,　we　design　and　conduct　experiments　of　attacking　ASSP　knapsacks　by　LLL　lattice　basis　reduction,　and　then　analyze　the　experimental　data.　Our　experiments　show　that　the　success　rate　varies　with　the　density　and　length　of　a　knapsack,　and　it　is　almost　0　when　the　density　is　larger　than　1.　Hence　we　can　conclude　that　the　JUNA　cryptosystem　can　resist　LLL　lattice　basis　reduction.　Our　experimental　data　in　the　paper　provides　the　factual　basis　for　our　further　research　on　the　security　of　JUNA.