[refactor](next) performance tuning (#3603)

Author: morningman

docs-next/query-acceleration/caching-intro.mdx

@@ -0,0 +1,37 @@
+---
+{
+    "title": "Caching",
+    "language": "en",
+    "description": "Navigate Apache Doris caching capabilities: SQL Cache for query result reuse, Condition Cache for repeated filter acceleration, and Data Cache for external table file caching."
+}
+---
+
+import GettingStartedCard from '@site/src/components/getting-started-card/getting-started-card';
+
+Apache Doris provides multiple layers of caching to accelerate queries: result-level caching reuses query outputs across identical SQL, segment-level caching reuses filter evaluations across queries, and file-level caching brings remote lakehouse data closer to the compute layer. Pick the cache that matches your workload pattern.
+
+## Result & Filter Cache
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="SQL Cache"
+    description="Cache full query results keyed by SQL text and table versions to skip repeated computation in T+1 and low-update scenarios."
+    link="sql-cache-manual"
+/>
+
+<GettingStartedCard
+    title="Condition Cache"
+    description="Cache the filtering result of a condition on each segment so subsequent queries with the same predicate can skip redundant scans and filtering."
+    link="condition-cache"
+/>
+</div>
+
+## External Table File Cache
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="Data Cache"
+    description="Cache files from HDFS and object storage on local disks to accelerate Hive, Iceberg, Hudi, and Paimon queries with cache warmup and admission control."
+    link="../lakehouse/data-cache"
+/>
+</div>

docs-next/query-acceleration/distinct-counts/intro.mdx

@@ -0,0 +1,31 @@
+---
+{
+    "title": "Distinct Counts",
+    "language": "en",
+    "description": "Navigate Apache Doris distinct count solutions: precise deduplication with BITMAP and approximate deduplication with HLL."
+}
+---
+
+import GettingStartedCard from '@site/src/components/getting-started-card/getting-started-card';
+
+Distinct count (deduplication) is one of the most resource-intensive operations in analytics. Apache Doris provides two purpose-built data types to replace `COUNT DISTINCT` with far lower memory and latency cost: choose **BITMAP** when you need exact results, and **HLL** when a 1%–2% error is acceptable in exchange for even smaller storage.
+
+## Precise Deduplication
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="BITMAP Precise Deduplication"
+    description="Replace COUNT DISTINCT with the BITMAP type for exact deduplication, faster queries, and lower memory and disk usage."
+    link="bitmap-precise-deduplication"
+/>
+</div>
+
+## Approximate Deduplication
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="HLL Approximate Deduplication"
+    description="Use HyperLogLog for high-cardinality UV and distinct-count workloads with controlled 1%–2% error and minimal storage footprint."
+    link="hll-approximate-deduplication"
+/>
+</div>

docs-next/query-acceleration/high-concurrency-intro.mdx

@@ -0,0 +1,31 @@
+---
+{
+    "title": "High Concurrency & Point Queries",
+    "language": "en",
+    "description": "Navigate Apache Doris capabilities for high-concurrency, low-latency workloads: primary-key point query optimization with row store and short-circuit execution, and dictionary tables for in-memory key-value lookups."
+}
+---
+
+import GettingStartedCard from '@site/src/components/getting-started-card/getting-started-card';
+
+Apache Doris is built for analytical scans, but with the right table design and execution paths it can also serve high-QPS point queries and key-value lookups. Use primary-key point query optimization for short-path retrieval on Unique tables, and dictionary tables to replace dimension-table joins with in-memory lookups.
+
+## Point Queries
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="High-Concurrency Point Query"
+    description="Enable row store, short-circuit execution, and PreparedStatement on Unique-key tables to push primary-key lookups to high QPS with low latency."
+    link="high-concurrent-point-query"
+/>
+</div>
+
+## Key-Value Lookup
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="Dictionary Table"
+    description="Pre-load dimension columns into memory and replace LEFT OUTER JOIN with dict_get function calls for fast key-value lookups in JOIN-heavy workloads."
+    link="dictionary"
+/>
+</div>

docs-next/query-acceleration/hints/hints-overview.md

@@ -0,0 +1,37 @@
+---
+{
+    "title": "Join Optimization",
+    "language": "en",
+    "description": "Navigate Apache Doris Join optimization techniques: Colocation Join, Distribute Hint for shuffle/broadcast, and Leading Hint for join order."
+}
+---
+
+import GettingStartedCard from '@site/src/components/getting-started-card/getting-started-card';
+
+Apache Doris adaptively optimizes most Join queries out of the box, but for performance-critical scenarios you can guide the planner with table colocation and hints. Start with Colocation Join to remove network shuffle for bucket-aligned tables, then use Distribute and Leading hints to fine-tune shuffle method and join order when the optimizer's choice is suboptimal.
+
+## Colocation
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="Colocation Join"
+    description="Co-locate data of related tables on the same BE so bucket-column joins run locally, eliminating cross-node shuffle for the most demanding multi-table joins."
+    link="colocation-join"
+/>
+</div>
+
+## Hint-Based Tuning
+
+<div className="cards-grid">
+<GettingStartedCard
+    title="Adjusting Join Shuffle Mode"
+    description="Use the [shuffle] and [broadcast] Distribute Hints to override the optimizer's data distribution choice for the right table in a Join."
+    link="tuning/tuning-plan/adjusting-join-shuffle"
+/>
+
+<GettingStartedCard
+    title="Reordering Join with Leading Hint"
+    description="Use Leading Hint to manually specify the join order of multiple tables — left-deep, right-deep, or bushy — when CBO does not pick the ideal plan."
+    link="tuning/tuning-plan/reordering-join-with-leading-hint"
+/>
+</div>