inconsistent runs for Picard Markduplicates.jar

[caswater]

I have two bam files for two samples (a.bam and b.bam), and I executed the same code to remove duplicates from the 2 files in the following way:

for A.bam
java -Xmx25g -jar ~/tools/picard-tools-1.84/MarkDuplicates.jar I=a.bam METRICS_FILE=dup_picard.txt O=a.rmdup.bam REMOVE_DUPLICATES=true VALIDATION_STRINGENCY=SILENT TMP_DIR=local_rmdup

for B.bam
java -Xmx15g -jar ~/tools/picard-tools-1.84/MarkDuplicates.jar I=b.bam METRICS_FILE=dup_picard.txt O=b.rmdup.bam REMOVE_DUPLICATES=true VALIDATION_STRINGENCY=SILENT TMP_DIR=local_rmdup

However, the output for the two samples are somewhat different -
for a.bam, it gives histogram values and also the last column of estimated library size in the METRICS_FILE ('dup_picard.txt'). But the information is all missing for the other bam file, and only reported the estimated duplication rate.

Can anyone help on this, and what's the problem for the inconsistent outputs?
Thanks