Data Tiering: Persistent Memory, NSE & Extension Nodes

A team expects persistent memory to speed up their write-heavy HANA workload. Why is that the wrong expectation, and what does the DRAM/PMEM split actually optimize?

A customer wants to add persistent memory to an existing HANA appliance with DRAM populated on only some channels. Which prerequisites block this, and which one is the easy-to-miss planning trap?

You run ALTER TABLE ... PERSISTENT MEMORY on a large partitioned HANA table during business hours. Which option avoids an immediate cost spike, and what does CASCADE change that the others don't?

You set PERSISTENT MEMORY ON at the table level but want one column excluded. Does HANA's level hierarchy allow that, and what does DEFAULT actually mean?

What is the one defining property of HANA Native Storage Extension (NSE) that distinguishes it from normal column loading, and why does it shrink the memory footprint?

You're deciding whether a large rarely-queried column should be COLUMN LOADABLE or PAGE LOADABLE in HANA NSE. What's the single trade-off, and how risky is getting it wrong?

Before manually marking columns page-loadable in HANA NSE, why run the NSE advisor first, and what does it base its recommendations on?

You install HANA with persistent memory enabled via the lifecycle manager. What is the default table behavior afterward, and why can that surprise you?

A standard HANA node uses a 50:50 data-to-work-area ratio. How far can an extension node relax that, and what single assumption makes the relaxation safe?

Why must a HANA extension node sit in its own worker group rather than the default group, and what breaks if it doesn't?

As you move HANA data from hot to cold tiers, which way do volume, price, and performance each move — and what does that mean for placing a dataset?

What does the Data Warehousing Foundation's Data Lifecycle Manager give you that manually issuing tier-change DDL does not?

Adopting HANA dynamic tiering, you must decide how a table maps to the disk-based extended store. What's the difference between extended tables and multistore tables?

Which HANA monitoring views tell you whether a table or column uses persistent memory, and where do you read total PMEM usage per NUMA node?

You have one standby node and want it to protect both the hot worker nodes and the extension node in HANA. Why doesn't that work by default, and how do you fix it?

Before running table redistribution in a scale-out HANA system with an extension node, what must you do first, and what goes wrong if you skip it?

Data aging and NSE both reduce HANA's memory footprint. What kind of mechanism is data aging, and what bounds where you can use it?

In HANA NSE, which views give buffer-cache statistics, and how do you list which tables are actually page-loadable?

What concrete operational pain does HANA persistent memory remove that DRAM causes, and how does it change your maintenance planning?

What is the one discriminator that separates warm data from cold data in HANA tiering — and why does it matter operationally?

After setting up an extension node and NSE, you need to verify placement. Which tool answers 'where is the extension node' versus 'is this table actually being paged'?

An extension node only holds warm data, so a colleague wants to build it on commodity hardware. Why is that wrong, and what actually defines an extension node?

How big is the HANA NSE buffer cache by default, what's a sane starting size for warm data, and what happens if you set both sizing parameters?

You're choosing a warm-data tier for an S/4HANA system and consider dynamic tiering. Why is that a mistake, and what should you use instead?

What's the smallest HANA configuration that can use an extension node, and does one hot node plus one extension node count as scale-out?

Which HANA parameter sets where persistent-memory storage lives versus which one turns it on for the whole database, and what must the location value line up with?

You UNLOAD a column in HANA expecting to reclaim memory, but usage barely drops. Why, and what actually frees it?

A colleague argues that since persistent memory survives shutdown, you can relax backups and system replication. Why is that dangerous?