I understand that each node is different and will have a different mempool (and size). But in looking at https://tradeblock.com/bitcoin/ it is showing a mempool size of around 20MB (http://bitcointicker.co/networkstats/ shows similar mempool size of 20MB). But https://blockchain.info/charts/mempool-size is showing a mempool of 103MB (https://jochen-hoenicke.de/queue/#24h showing similar 100+ MB size). Neither site explains how they configure their mempool. tradeblock.com and blockchain.info are referenced from many sources - do you know what can explain this big difference in how they report mempool size? thanks
Nodes have different policy about accepting transactions to mempool. Some nodes accept all transactions, others accept only transactions with a fee greater than 10 satoshi per byte for example.
Nodes have different policy about droping transaction from mempool. Some nodes drop one-day-old transactions, other nodes drop three-days-old transactions.
Nodes have different policy - this is the answer.
Blockchain.info is the most known and the biggest blockchain explorer. That's why they have the most number of peers (compared to other block explorers) and hence they have the biggest mempool.
You can see how many peers they have here (after they fix it).