On this page本页目录
What Is Data Virtualization Software?什么是数据虚拟化软件？
For the full topic map and the neighboring methods that support this workflow, continue with the federated queries and data virtualization guide.
如需查看完整主题结构以及支撑本流程的相邻方法，请继续阅读联邦查询与数据虚拟化指南。
Data virtualization software is the executable layer that connects distributed sources, presents governed logical data objects, plans queries across those sources, and returns results without requiring every dataset to be centralized first. A suitable product must work with your actual source versions, identities, queries, consumers, service objectives, and operating constraints—not merely list the right category features.
数据虚拟化软件是连接分散来源、呈现受治理逻辑数据对象、跨来源规划查询并返回结果的可执行层，不要求先集中每个数据集。合适产品必须适用于你的真实源版本、身份、查询、使用者、服务目标和运维约束，而不只是列出正确的类别功能。
This page addresses commercial investigation and implementation evaluation. A companion data virtualization architecture guide in this delivery set explains the underlying concept but is not yet deployed. Here, the task is to convert requirements into a shortlist, run comparable proof-of-concept tests, expose disqualifying gaps, and document a decision.
本页回答商业调研和实施评估意图。本交付集中另有一篇解释底层概念的数据虚拟化架构指南，但它尚未部署；这里的任务是把需求转化为候选名单，运行可比较的概念验证测试，暴露足以淘汰候选的缺口，并记录决策。
Do not confuse data virtualization software with data visualization software, virtual-machine software, database cloning, test-data virtualization, storage virtualization, or a catalog that cannot execute data access. The defining capability is governed logical access with an executable query or service path. “Real time” should be tested as a measurable freshness and latency contract, never accepted as a universal promise.
不要把数据虚拟化软件与数据可视化软件、虚拟机软件、数据库克隆、测试数据虚拟化、存储虚拟化或无法执行数据访问的目录混为一谈。其定义性能力是带有可执行查询或服务路径的受治理逻辑访问。“实时”应作为可测量的新鲜度与延迟契约进行测试，不能被当作普遍承诺。
Start with a Data Virtualization Software Selection Brief先编写数据虚拟化软件选型简报
A product demonstration is easy to optimize for. A selection brief makes the evaluation answer your operating reality. Record the following before inviting vendors or installing an engine:
产品演示很容易被专门优化。选型简报能迫使评估回答你的真实运行条件。在邀请供应商或安装引擎前，应记录以下内容：
- Business outcomes: name the decisions, applications, reports, or data products that need access; assign owners and failure impact.业务结果：明确需要访问的决策、应用、报表或数据产品，并指定负责人和失败影响。
- Source inventory: list products, editions, versions, regions, network paths, authentication, schemas, volume, change rate, query limits, maintenance windows, and accountable owners.来源清单：列出产品、版本、区域、网络路径、身份验证、Schema、规模、变化率、查询限制、维护窗口和责任人。
- Consumer contracts: capture SQL dialects, JDBC or ODBC needs, APIs, BI tools, applications, concurrency, result size, freshness, availability, and acceptable partial-result behavior.使用契约：记录SQL方言、JDBC或ODBC需求、API、BI工具、应用、并发、结果大小、新鲜度、可用性和可接受的部分结果行为。
- Governance: define identity propagation, row and column rules, masking, residency, encryption, audit retention, lineage, approval, export, and deletion obligations.治理：定义身份传递、行列规则、遮蔽、驻留、加密、审计保留、血缘、批准、导出和删除义务。
- Service objectives: set workload-specific thresholds for correctness, p50/p95/p99 latency, throughput, queue time, source pressure, recovery, and cost.服务目标：按负载设定正确性、p50/p95/p99延迟、吞吐、排队、源压力、恢复和成本阈值。
- Constraints and exit: state deployment boundaries, skills, support hours, procurement rules, lock-in tolerance, portability, data egress, and required handoff artifacts.约束与退出：说明部署边界、技能、支持时段、采购规则、锁定容忍度、可移植性、数据出站和必需交接物。
Gate before shortlisting: if the team cannot provide representative queries, authorized test data, source monitoring, and measurable pass/fail thresholds, pause procurement. The POC would otherwise reward presentation quality rather than production fit.
候选前置门槛：如果团队无法提供代表性查询、获授权测试数据、源监控以及可测量的通过/失败阈值，应暂停采购，否则POC奖励的是演示质量，而不是生产适配度。
Core Data Virtualization Software Features to Verify需要验证的数据虚拟化软件核心功能
Verify supported versions, authentication, TLS, data types, statistics, predicate and aggregation pushdown, cancellation, metadata refresh, error mapping, and upgrade policy.
验证支持版本、身份验证、TLS、数据类型、统计信息、谓词与聚合下推、取消、元数据刷新、错误映射和升级策略。
Test reusable views, relationships, calculations, business terms, schema contracts, lineage, ownership, versioning, impact analysis, and semantic consistency.
测试可复用视图、关系、计算、业务术语、Schema契约、血缘、所有权、版本、影响分析和语义一致性。
Inspect cost estimates, join placement, pushdown, parallelism, dynamic filtering, memory, spill, cache, result reuse, workload groups, quotas, and explain plans.
检查成本估算、连接位置、下推、并行、动态过滤、内存、暂存、缓存、结果复用、负载组、配额和执行计划。
Prove single sign-on, source identity or service-account behavior, row and column policies, masking, secrets, encryption, audit, residency, and policy consistency.
证明单点登录、源身份或服务账户行为、行列策略、遮蔽、密钥、加密、审计、驻留和策略一致性。
Confirm SQL dialect, JDBC/ODBC, APIs, BI compatibility, prepared statements, result pagination, metadata discovery, service publication, and client error behavior.
确认SQL方言、JDBC/ODBC、API、BI兼容、预编译语句、结果分页、元数据发现、服务发布和客户端错误行为。
Require metrics, logs, traces, plan history, lineage, alerting, high availability, backup, disaster recovery, patching, rollback, automation, and support evidence.
要求指标、日志、追踪、计划历史、血缘、告警、高可用、备份、灾难恢复、补丁、回滚、自动化和支持证据。
A “yes” in a feature matrix is not evidence. Record the scope and limitation: which connector, version, operation, identity, deployment mode, license tier, and failure condition were tested. Trino’s official documentation is a useful reminder that pushdown support is connector- and source-specific; even a generally supported optimization may not apply to a particular function or query shape.
功能矩阵中的“是”不是证据。应记录测试范围和限制：测试了哪个连接器、版本、操作、身份、部署模式、许可层级和失败条件。Trino官方文档提醒我们，下推支持取决于连接器和来源；即使某项优化通常受支持，也未必适用于特定函数或查询形态。
Build a Weighted Data Virtualization Software Scorecard建立加权数据虚拟化软件评分表
Weight criteria before seeing vendor results. The weights below are a hypothetical example, not a universal prescription; adjust them to your risk and workload. Use a 0–5 evidence score and multiply by the approved weight. A roadmap statement should not receive the same score as a repeatable test.
应在看到供应商结果前确定权重。下列权重是假设示例，不是普遍规则；应按你的风险与负载调整。使用0–5证据分数并乘以获批权重。路线图声明不应获得与可重复测试相同的分数。
| Criterion标准 | Example weight示例权重 | Required evidence必需证据 |
|---|---|---|
| Connector and semantic correctness连接器与语义正确性 | 20 | Version matrix, type/null/time-zone cases, pushdown plans, cancellation, schema-change tests版本矩阵、类型/null/时区用例、下推计划、取消和Schema变化测试 |
| Performance and source protection性能与源保护 | 20 | Cold/warm percentiles, concurrency, transferred bytes, source CPU/I/O/connections, limits and recovery冷热分位数、并发、传输字节、源CPU/I/O/连接、限制与恢复 |
| Security and governance安全与治理 | 20 | Allowed and denied identities, row/column policies, masking, audit, lineage, secrets and residency evidence允许与拒绝身份、行列策略、遮蔽、审计、血缘、密钥和驻留证据 |
| Reliability and operations可靠性与运维 | 15 | Dependency failure, timeout, circuit breaker, HA, backup/restore, upgrade, rollback and on-call runbooks依赖失败、超时、熔断、高可用、备份恢复、升级、回滚和值班手册 |
| Consumer and developer experience使用者与开发体验 | 10 | Real client compatibility, discoverability, versioning, debugging, automation and deployment workflow真实客户端兼容、可发现性、版本、调试、自动化和部署工作流 |
| Economics, support and exit经济性、支持与退出 | 15 | Measured consumption, labor model, license terms, support SLA, portability, export and replacement plan实测消耗、人员模型、许可条款、支持SLA、可移植性、导出和替代计划 |
Add non-negotiable gates outside the weighted total. For example: a residency violation, missing required connector version, incorrect row policy, uncontrolled load on a transaction system, or silent partial result should fail the candidate even if its overall score is high.
还应在加权总分之外设置不可妥协门槛。例如，违反驻留要求、缺少必需连接器版本、行策略错误、对交易系统产生不可控负载，或静默返回部分结果，都应淘汰候选，即使其总分很高。
Design a Defensible Data Virtualization Software POC设计可辩护的数据虚拟化软件POC
A useful POC is small enough to finish and broad enough to reveal the architecture. Include six workload families:
有用的POC应足够小以便完成，又足够广以暴露架构。至少包含六类负载：
- Selective lookup: indexed filters, small result, parameterized execution, and permission checks.选择性查找：索引过滤、小结果、参数化执行和权限检查。
- Pushdown proof: predicates, projections, aggregations, limits, functions, and an unsupported variant; inspect both virtual and remote plans.下推证明：谓词、投影、聚合、限制、函数以及一个不受支持变体；检查虚拟与远程计划。
- Cross-source join: realistic skew, nulls, duplicates, type differences, time zones, and one source that cannot execute the join.跨源连接：真实倾斜、null、重复、类型差异、时区以及一个无法执行连接的来源。
- Repeated analytical workload: cold and warm runs, cache or result reuse, concurrency, queues, memory, spill, cancellation, and cost.重复分析负载：冷热运行、缓存或结果复用、并发、队列、内存、暂存、取消和成本。
- Governance matrix: permitted and denied users, service accounts, row and column rules, masking, exports, cache isolation, audit, and lineage.治理矩阵：允许与拒绝用户、服务账户、行列规则、遮蔽、导出、缓存隔离、审计和血缘。
- Failure and change: timeout, unavailable source, credential expiry, quota, network interruption, schema change, bad statistics, upgrade, rollback, and recovery.失败与变化：超时、来源不可用、凭据过期、配额、网络中断、Schema变化、错误统计、升级、回滚和恢复。
Control the comparison: freeze test data or record expected snapshots, retain source metrics, repeat runs, and preserve configuration. If a vendor tunes a query, require the change to be documented and give the same opportunity to every candidate.
控制比较：冻结测试数据或记录预期快照，保留源指标，重复运行并保存配置。如果供应商调优查询，必须记录变更，并向每个候选提供同等机会。
Hypothetical Example: Available-to-Promise Inventory假设示例：可承诺库存
Hypothetical scenario: a retailer wants an available-to-promise view combining ERP stock, warehouse reservations, e-commerce orders, and a small product reference. The numbers below are illustrative, not observed InfiniSynapse or vendor results.
假设场景：某零售商希望建立可承诺库存视图，组合ERP库存、仓库预留、电商订单和小型商品参考。下列数字仅为示例，并非InfiniSynapse或任何供应商的实测结果。
| Test测试 | Hypothetical threshold假设阈值 | Decision evidence决策证据 |
|---|---|---|
| Business correctness业务正确性 | Exact match on 60 approved fixtures, including null, duplicate, cancellation, and late-reservation cases60个获批样例完全匹配，包括null、重复、取消和延迟预留场景 | Expected and actual rows, rule version, source snapshot, discrepancy owner预期与实际行、规则版本、源快照、差异负责人 |
| Interactive query交互查询 | p95 under 4 seconds at 20 concurrent users; no silent partial result20并发用户时p95低于4秒；不得静默返回部分结果 | Percentiles, queue time, plans, transferred bytes, errors and cancellations分位数、排队时间、计划、传输字节、错误和取消 |
| ERP protectionERP保护 | No agreed CPU/connection/lock threshold breach during the test window测试窗口内不突破约定CPU、连接或锁阈值 | ERP telemetry before, during and after load; enforced query limits and recovery负载前中后的ERP遥测、已执行查询限制和恢复 |
| Outage behavior中断行为 | Unavailable source identified within 15 seconds; stale or partial behavior follows the approved policy15秒内识别不可用来源；陈旧或部分行为遵循获批策略 | Error contract, alert, trace, audit event, fallback decision and recovery time错误契约、告警、追踪、审计事件、回退决定和恢复时间 |
Suppose a candidate meets correctness and security but exceeds the ERP load gate when a cross-source join cannot be pushed down. The defensible result may be conditional approval: replicate only the small operational subset to an analytical replica, retain virtual access for current reference data, and retest. That is better than hiding the failure inside a blended score.
假设某候选满足正确性和安全要求，却因跨源连接无法下推而超过ERP负载门槛。可辩护结果可能是条件批准：只把小型运营子集复制到分析副本，对当前参考数据保留虚拟访问，然后重新测试。这优于把失败隐藏在综合分数中。
Use InfiniSynapse for a Bounded Multi-Source Analysis Task使用InfiniSynapse完成有界多源分析任务
InfiniSynapse’s public site describes direct connections to supported databases and multi-source joint analysis without requiring a complex migration first. With approved access, definitions, keys, grain, privacy rules, freshness expectations, and test examples, use the app to explore a bounded question across relevant supported connections.
InfiniSynapse官网说明其支持直接连接受支持数据库和多源联合分析，无需先完成复杂迁移。准备获批访问、定义、键、粒度、隐私规则、新鲜度期望和测试样例后，可使用应用跨相关受支持连接探索有界问题。
This related workflow does not establish InfiniSynapse as general-purpose data virtualization software. We do not claim it publishes reusable virtual SQL services, exposes optimizer or pushdown controls, manages enterprise caches, replaces ETL or a warehouse, or supplies procurement infrastructure controls. Verify visible task support and keep architecture, credentials, approvals, load protection, contracts, and deployment in governed systems.
这一相关工作流不证明InfiniSynapse是通用数据虚拟化软件。本页不声称它会发布可复用虚拟SQL服务、暴露优化器或下推控制、管理企业缓存、取代ETL或数仓，或提供采购基础设施控制。应验证可见任务支持，并把架构、凭据、批准、负载保护、合同和部署留在受治理系统中。
Bring approved connections, a bounded question, definitions, keys, grain, privacy constraints, freshness expectations, and trusted examples. Run the supported analysis, then validate it against agreed evidence.
请准备获批连接、有界问题、定义、键、粒度、隐私约束、新鲜度期望和可信样例。运行受支持分析，再依据约定证据验证。
Analyze approved connected sources分析获准的已连接来源Data Virtualization Software FAQ数据虚拟化软件常见问题
What is data virtualization software?
什么是数据虚拟化软件？
Data virtualization software creates and operates a logical access layer over distributed sources. Depending on the product, it may provide connectors, virtual views, query planning and pushdown, security policies, metadata, caching, observability, and SQL or API consumption without requiring every dataset to be copied first.
数据虚拟化软件在分散来源之上创建并运行逻辑访问层。不同产品可能提供连接器、虚拟视图、查询规划与下推、安全策略、元数据、缓存、可观测性以及SQL或API消费，而不要求先复制每个数据集。
What features should data virtualization software have?
数据虚拟化软件应具备哪些功能？
The required features depend on the workload, but a serious evaluation should test source-specific connectors, semantic correctness, pushdown and distributed planning, identity and policy enforcement, lineage, workload controls, monitoring, failure behavior, versioning, and supported consumer interfaces.
所需功能取决于负载，但严肃评估应测试来源专属连接器、语义正确性、下推与分布式规划、身份与策略执行、血缘、负载控制、监控、失败行为、版本管理和受支持的消费接口。
How do you compare data virtualization software vendors?
如何比较数据虚拟化软件供应商？
Use the same representative queries, source versions, identities, failure cases, concurrency, and acceptance thresholds for every candidate. Score observed evidence separately from roadmap promises, and include operating effort, licensing, support, portability, and exit costs.
对每个候选使用相同的代表性查询、源版本、身份、失败场景、并发和验收阈值。把已观察证据与路线图承诺分开评分，并纳入运维投入、许可、支持、可移植性和退出成本。
Does data virtualization software replace ETL or a data warehouse?
数据虚拟化软件会取代ETL或数据仓库吗？
Usually no. Virtual access suits current, selective, or movement-constrained workloads, while ETL, ELT, warehouses, and lakehouses remain useful for durable history, repeatable snapshots, heavy reuse, isolation, and predictable high-concurrency performance. Many production designs use both.
通常不会。虚拟访问适合当前、选择性或移动受限的负载；ETL、ELT、数仓和湖仓仍适合持久历史、可重复快照、重度复用、隔离以及可预测的高并发性能。许多生产设计会同时使用两者。
How should a data virtualization software POC be tested?
数据虚拟化软件POC应该如何测试？
Run a small set of business-correct workloads covering pushdown, cross-source joins, concurrency, source protection, permissions, failure and recovery, schema change, cache freshness, and cost. Define pass, conditional pass, and fail thresholds before vendors tune the tests.
运行一组小而完整的业务负载，覆盖下推、跨源连接、并发、源保护、权限、失败与恢复、Schema变化、缓存新鲜度和成本。在供应商调优测试之前先定义通过、条件通过和失败阈值。
Official Sources and Verification Notes官方来源与验证说明
Capabilities vary by product and connector. Use these official sources to define test questions, not to assume feature parity or production fit.
能力会因产品和连接器而变化。请用这些官方资料定义测试问题，不要据此假定功能等同或生产适配度。