SAP HANA 基础架构简介（三）

Index Server –JoinEngine

• 当query 包含JOIN or aggregate expressions 时使用
  • e.g. COUNT, MIN, MAX, SUM, AVG, …
• 在Join Engine 之前, the SQL Layer 创建一个Join View
  • Then the query is executed on the view
• Join View (≡ Attribute View)
• Statistics of Join Engine:

SELECT * FROM "SYS"."M_DEV_JOINENGINE_"

Index Server –OLAP Engine

• OLAP View ≡ Analytical View
• OLAP Views can be created via SQL command or via the modeler
• WITH HINT(OLAP_PARALLEL_AGGREGATION) at the end of a query can be used to execute the query on the OLAP Engine
  • A Join View is created in all cases
  • OLAP Engine tries to create an OLAP View and execute the query on it
  • If it’s impossible because the OLAP Engine cannot handle the query, a Join View is used

Index Server –RowStore / ptime

• 行存储数据
• 表存在于共享内存段
• 用于OnLine-Transaction-Processing
  • Response time < 1 msec
  • 简单的 SELECTqueries 比Column Store执行更好
• 每个内存中的表使用 a linked list of pages (16k each)
  • Pages consist of a header and of slots that contain the table records
  • Memory is not allocated page by page to avoid memory fragmentation
    • Instead segments of 64 MB are allocated
    • A segment is freed only when all it’s pages are empty
• 无数据压缩
• 多节点系统中只存在于Master中

Index Server –Column Store / trex

• 列存储数据
• 数据存在于Heap (dynamic memory)
• 使用OnLine-Analytical-Processing
  • Response time < 100 msec
• 优化复杂的 analytical queries and aggregation
  • Performs comparatively poor for a high number of small SELECTqueries
• Tables 被分裂到partitions
  • Split is done in a way that each partition contains a different set of rows of the table
  • 只有 Column Store 支持 table splitting
• 多几点系统表可以分布是存储
• 数据压缩
  • 目标: 存储更多数据到主存
  • 问题: 写操作到压缩数据会花时间
    • 需要重组存储结构并重新计算压缩
    • 因此写操作并不直接编辑压缩的数据
• 写入Delta Storage的改变
  • Delta Storage 只存在内存中
  • Only Delta Log entries are written to the Persistence Layer when entries are inserted into the delta storage
  • After a HANA restart the Delta Log entries are replayed to restore the Delta Storage
  • Main part of the data (→ compressed data) is called Main Storage
  • Changes are taken over from Delta Storage to Main Storage asynchronously during Delta Merge

PersistenceLayer

• 是必要的，因为主内存不够稳定
• 存储数据到persistent disk volumes
• Volumes are organized in pages
  • Page sizes between 4k and 16Msupported
• 提供备份还原功能

XSEngine

• Application server 允许clients 访问HANA 通过HTTP
  • Allows running JavaScript web applications on HANA server side that can use the DB
• 可以嵌入Index Server 1849775或作为外部服务
• 外部XSEngine服务是一个没有表的Index Server
  • All incoming queries are forwarded to the HANA landscape
• Xsa:<application_name> for application trace