JanusGraphTransaction tx = janusGraph.newTransaction();
for(String key:keyList){

    String[] arry = key.split("#");
    String instructionId = arry[0];
    String transactionId = arry[1];
    long updateStart = System.currentTimeMillis();
    tx.traversal().V().has(NODE_TYPE_INSTRUCTION, ID_TYPE_INSTRUCTION, instructionId).out(EDGE_IT).has(ID_TYPE_TRANSACTION, transactionId).next().property(PARAM_0, CHANGED_PARAM_VAL);
    long timeToUpdate = System.currentTimeMillis() - updateStart;
    LOGGER.info("Updated in " + timeToUpdate +" " + instructionId + ":" +  transactionId);
}
tx.commit();

In the log for updates i see that some updates take around 1-2 seconds while some updates are done within 10 milliseconds.

Can you please clarify me the reason for time difference?

Is there any better way to do bulk updates in Java api?

What i assumed was when i use a transaction boundary as above the updates will not go to back-end till i commit. But if that is so there can't be any time difference shown.
So what is the behavior here?

Thanks
Dilan.