[doc] Add read performance doc.

docs/content/maintenance/read-performance.md

@@ -0,0 +1,68 @@
+---
+title: "Read Performance"
+weight: 2
+type: docs
+aliases:
+- /maintenance/read-performance.html
+---
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements.  See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership.  The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License.  You may obtain a copy of the License at
+
+  http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied.  See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+# Read Performance
+
+There are many ways to improve read performance when reading Paimon data with Flink.
+
+## Acceleration In Flink client
+When use Flink batch job scan Paimon table, if you don't add option `scan.parallelism`, Paimon will infer the parallelism by read manifest files which take lots of time. So you can read Paimon table with the `scan.parallelism` table property.
+
+## Read Compact Snapshot
+You can read snapshot which commitKind = Compact, Compared to read other commitKind snapshot, reading this snapshot requires fewer merge operations, resulting in higher performance.
+
+## Modify Compaction Config
+You can reduce the option value of configuration `num-sorted-run.compaction-trigger` for higher performance. But this is a trade-off between writing and query performance.
+
+## Modify DataFile's Compression Rates
+You can add option `file.compression.zstd-level` to adjust DataFile's compression rates, the smaller the value, the higher the read performance.
+
+## DeletionVector + Vectorized Computing Engine
+You can set `deletion-vectors.enabled` = true, and use a computing engine with vectorized ability to read Paimon table. This way can fully utilize the SIMD capability of the CPU to accelerate data processing.
+
+# Avoid OOM
+Reading data often leads to OOM due to various reasons. Below are several methods to reduce the probability of OOM.
+
+## Reduce The Concurrency Of Reading Manifest
+The parallelism of scanning manifest files, default value is the size of cpu processor. We can control it by `scan.manifest.parallelism`, the smaller the value, the lower probability of OOM.
+
+## Reduce The Stats In DataFileMeta
+If your Paimon table has lots of columns, and you just want field B to use dense stat, other columns close stats. You can control this by follow options.
+
+`metadata.stats-dense-store` = true
+
+`metadata.stats-mode` = none
+
+`fields.b.stats-mode` = truncate(16)
+
+## Reduce The Record In Memory
+If your Paimon table has some lager records which will occupy a lot of memory. Your can reduce the option `read.batch-size`, the default value is 1024.
+
+## Optimize Memory When MOR
+When merge on read in a sortedRun which include lots of DataFile, too many readers will consume too much memory and causing OOM.
+Your can add option `sort-spill-threshold` to spill some DataFile before MOR.
+
+