The suggested idea I was replying to is to make all dust TXs invalid by some nodes. I suggested a compromise by keeping them in secondary storage for full nodes, and in a separate Merkle Tree for bridge servers.
-In bridge servers they won't increase any worstcase, on the contrary this will enhance the performance even if slightly.
-In full nodes, and since they will usually appear in clusters, they will be fetched rarely (either by a dust sweeping action, or a malicious attacker)
In both cases as a batch
-To not exhaust the node with DoS(as the reply mentioned)one may think of uploading the whole dust partition if they were called more than certain threshold (say more than 1 Tx in a block)  
-and then keep them there for "a while", but as a separate partition too to exclude them from any caching mechanism after that block.
-The "while" could be a tuned parameter.
-Take care that the more dust is sweeped, the less dust to remain in the UTXO set; as users are already much dis-incentivised to create more.
.
Thanks for allowing the reply

On Thu, Oct 7, 2021, 16:43 ZmnSCPxj <ZmnSCPxj@protonmail.com> wrote:


> I don't know what brings up sorting here, unless as an example.

Yes, it is an example: quicksort is bad for network-facing applications because its ***worst-case behavior*** is bad.
Bitcoin is a network-facing application, and similarly, ***worst-case behavior*** being bad is something that would strongly discourage particular approaches.
Your proposal risks bad ***worst-case behavior***.

> Anyways, I was comparing to rejecting them completely, not to keeping them in one set. In addition, those dust sweep Transactions will probably be a dust sweep and thus contain so many inputs which "maybe" makes 1-one disk visit  to fetch all their hashes at once, 2-from a smaller subset with max size 5-10% the UTXO set, justifiable.

Do not consider the ***average case*** where a block is composed of only a few dust sweep transactions and most transactions are normal, non-dust-sweep transactions.

Instead, consider the ***worst case*** where ***all*** transactions in a block are dust sweep transactions, because that is what attackers will use.

Regards,
ZmnSCPxj