評估部門共同作業環境的效能和容量需求 (SharePoint Server 2013)Estimate performance and capacity requirements for divisional collaboration environments (SharePoint Server 2013)

摘要:使用測試結果和建議來評估 SharePoint Server 2013 的部門共同作業環境的效能和容量需求。Summary: Use test results and recommendations to estimate performance and capacity requirements for a divisional collaboration environment for SharePoint Server 2013.

本文說明關於效能和容量規劃根據 SharePoint Server 2013 的部門共同作業解決方案的指引。本文隨附的下列資訊:This article describes guidance about performance and capacity planning for a divisional collaboration solution that is based on SharePoint Server 2013. The following information is included in this article:

  • 測試實驗室環境規格,例如硬體、伺服器陣列拓撲和設定Test lab environment specifications, such as hardware, farm topology, and configuration

  • 測試伺服器陣列工作負載,以及產生該測試負載的資料集The test farm workload and dataset that generated the test load

  • 測試結果和分析,其中示範並說明在特定規模點的負載下,輸送量、延遲與硬體需求的趨勢。Test results and analysis that demonstrate and explain trends in throughput, latency, and hardware demand under load at specific scale points.

請使用本文中的資訊,了解在一般負載與尖峰負載下的案例特性,以及當陣列伺服器向外延展時,效能趨勢如何改變。本文也可協助您評估自己規劃的架構該以什麼為適當的起點,並了解當您在進行相關規劃,以將尖峰負載下的效能維持在可接受的程度時,應該考慮的事項。Use the information in this article to understand the characteristics of the scenario under both normal and peak loads, and how performance trends change when farm servers are scaled out. This article can also help you estimate an appropriate starting point for your planned architecture, and understand things to consider when you develop a plan to maintain acceptable levels of performance under peak load.

簡介Introduction

本文概述如何向外延展伺服器中的 SharePoint Server 2013 部門共同作業解決方案。部門共同作業解決方案是具有較少的電腦中的共同作業活動的相關比企業共同作業解決方案的 SharePoint Server 2013 部署。本文假設部門內有 1000 到 10000 位員工企業組織。This article outlines how to scale out servers in a SharePoint Server 2013 divisional collaboration solution. A divisional collaboration solution is a SharePoint Server 2013 deployment that has fewer computers involved in the collaborative activities than an enterprise collaboration solution. This article assumes that a division is an organization inside an enterprise that has 1,000 to 10,000 employees.

不同的案例有不同的需求。因此,如果本指南未提到您的情況,請另行以自家的硬體和環境進行測試。如果您規劃的設計和工作負載跟本文所述的環境很像,可以直接根據本文來預測您向上與向外延伸環境後會達到的效能。Diverse scenarios will have different requirements. Therefore, supplement this guidance with additional testing on your own hardware and in your own environment. If your planned design and workload resembles the environment described in this article, you can draw conclusions about the performance to expect when you scale your environment up and out.

重要

本文所示的測試結果是在實驗室中產生,過程中以嚴謹控制的條件,以工作負載、資料集與架構來模擬實際執行環境。雖然這些測試經過十分仔細的設計,但在測試實驗室中得到的效能特性永遠不會和實際執行環境中的行為相同。這些測試結果無法代表實際執行伺服器陣列的效能與容量特性。不過,這些測試結果可顯示出在輸送量、延遲與硬體需求方面觀察到的趨勢。運用這些對於所觀察到資料的分析,您將更能規劃容量並管理伺服器陣列。Test results in this article were produced in a test lab that used a workload, dataset, and architecture to simulate a production environment under highly controlled conditions. While we carefully designed these tests, the performance characteristics of a test lab are never the same as the behavior of a production environment. These test results do not represent the performance and capacity characteristics of a production farm. Instead, the test results demonstrate observed trends in throughput, latency, and hardware demand. Use the analysis of the observed data to help you plan capacity and manage your own farm.

閱讀本文之後,您將了解:As you read this article, you'll learn about the following:

  • 規格的硬體、 拓撲及設定Specifications, which include hardware, topology, and configuration

  • 工作量,包括需求的伺服器陣列、 使用者數目及使用狀況特性的分析The workload, which includes an analysis of the demand on the farm, the number of users, and usage characteristics

  • 資料集,例如資料庫大小和內容類型The dataset, such as database sizes and content types

  • 測試結果與分析向外延展網頁伺服器Test results and analysis to scale out web servers

閱讀本文之前,請閱讀下列文章,以確認您了解SharePoint 2013 的軟體界限及限制SharePoint Server 2013 中的容量管理的重要概念。Before you read this article, read the following articles to make sure that you understand the key concepts behind capacity management in Software boundaries and limits for SharePoint 2013SharePoint Server 2013.

詞彙Glossary

下表包含可以在本文中找到之關鍵字詞的定義。The following list contains definitions for key terms found in this article.

  • RPS:每秒要求數目。RPS 為伺服器陣列或伺服器會收到一秒的要求數目。這是常見的伺服器與伺服器陣列負載度量單位。RPS: Requests per second. RPS is the number of requests that a farm or server receives in one second. This is a common measurement of server and farm load.

    注意

    要求與頁面載入不同。一個頁面包含多個元件,當瀏覽器載入頁面時,每個元件都會建立一或多個要求。載入單一頁面就會建立多個要求。驗證檢查與使用極少量資源的事件一般不會計算在 RPS 度量中。Requests differ from page loads. A page contains several components, each of which creates one or more requests when a browser loads the page. A single page load creates several requests. Normally, authentication checks and events that use insignificant resources are not counted in RPS measurements.

  • 綠色區域:綠色區域代表一組已定義下正常作業的情況,最多個預期每日的尖峰負載的負載特性。在這個範圍中運作的伺服器陣列應該能夠維持回應時間與介於參數可接受的延遲。Green Zone: Green Zone represents a defined set of load characteristics under normal operating conditions, up to expected daily peak loads. A farm that operates in this range should be able to sustain response times and latency that are within acceptable parameters.

    在此狀態下,伺服器可維持下列準則︰This is the state at which the server can maintain the following set of criteria:

    • 至少 75% 的要求在伺服器端的延遲少於 1 秒。The server-side latency for at least 75% of the requests is less than 1 second.

    • 所有的陣列伺服器維持平均 CPU 使用率在 60% 以下。All farm servers maintain an average CPU utilization of less than 60%.

      注意

      我們實驗室環境沒有執行使用中的搜尋編目。讓我們保留接近 50%的 CPU 使用率或較低齊備搜尋編目負載為 10%的資料庫伺服器。本範例假設 SQL 伺服器資源管理員在生產環境中用來限制搜尋編目負載為 10 %cpu。Our lab environment did not run an active search crawl. So we kept the database server close to 50% CPU utilization or lower to reserve 10% for the search crawl load. This assumes SQL Server Resource Governor is used in production to limit Search crawl load to 10% CPU.

    • 失敗率低於 0.01%。Failure rate is less than 0.01%.

  • 紅色區域 (Max):紅色區域代表一組已定義的下運作情況的尖峰負載特性。紅色區域在伺服器陣列的錯誤率非常高暫時性的資源需求愈多,它可以維持僅針對有限的段前失敗並發生其他效能和可靠性的問題。Red Zone (Max): Red Zone represents a defined set of load characteristics under peak operating conditions. At Red Zone, the farm experiences very high transient resource demands that it can sustain only for limited periods before failures and other performance and reliability issues occur.

    在此狀態下,伺服器可在有限的持續時間內維持下列準則︰This is the state at which the server can maintain the following set of criteria for a limited duration:

    • 啟用 HTTP 要求節流功能,但不傳回 503 錯誤 (伺服器忙碌中)。HTTP request throttling feature is enabled, but no 503 errors (Server Busy) are returned.

    • 失敗率低於 0. 1%。Failure rate is less than 0. 1%.

    • 至少 75% 的要求在伺服器端的延遲少於 3 秒。The server-side latency is less than 3 seconds for at least 75% of the requests.

    • 所有的陣列伺服器 (資料庫伺服器除外) 維持平均 CPU 使用率約在 90% 以下。All farm servers (excluding database servers) maintain an average CPU utilization of less than approximately 90%.

    • 資料庫伺服器的平均 CPU 使用率約在 50% 以下,以容許足夠的額外負荷保留給搜尋編目負載。Database server average CPU utilization is less than approximately 50%, which allows for ample overhead to be reserved for the Search crawl load.

  • AxBxC (圖表示法):網頁伺服器、 應用程式伺服器及資料庫分別在伺服器陣列伺服器的數目。例如,10 x 1 x 1 表示此環境中有 10 個網頁伺服器、 1 部應用程式伺服器及 1 的資料庫伺服器。AxBxC (Graph notation): The number of web servers, application servers, and database servers respectively in a farm. For example, 10x1x1 means that this environment has 10 web servers, 1 application server, and 1 database server.

  • MDF 和 LDF:SQL Server 實體檔案。如需詳細資訊,請參閱檔案與檔案群組架構MDF and LDF: SQL Server physical files. For more information, see Files and Filegroups Architecture.

概觀Overview

本節提供延展方法與測試方法的概觀。This section provides an overview of our scaling approach and test methodology.

延展方法Scaling approach

本節說明我們在延展實驗室環境時所採用的方法。此方法可讓您找到最適合您工作負載的設定︰This section describes the approach that we took to scale our lab environment. This approach will enable you to find the best configuration for your workload:

  • 我們向外延展網頁伺服器,直到使用四部網頁伺服器為止。每部伺服器都執行分散式快取服務。We scaled out the web servers until four web servers were being used. Each server runs the Distributed Cache Service.

  • 我們新增一部執行分散式快取服務的專用伺服器。We added a dedicated server that runs the Distributed Cache Service.

  • 我們停用網頁伺服器上的分散式快取服務。We disabled the Distributed Cache Service on the web servers.

  • 我們向外延展額外的網頁伺服器,直到測試範圍的最大值為止。We scaled out additional web servers to the maximum for the scope of testing.

方法與測試備註Methodology and test notes

因為本文包含測試實驗室環境的結果,所以我們能夠控制某些因素來顯示此工作負載下效能的特定層面。此外,實驗室環境省略了實際執行環境的某些要素 (列於以下清單中),以簡化測試的額外負荷。Because this article contains results from a test lab environment, we could control certain factors to show specific aspects of performance for this workload. In addition, certain elements of the production environment, which are in the following list, were left out of the lab environment to simplify the overhead of testing.

注意

我們建議您將這些要素包括在實際執行環境中。We recommend that you include these elements in production environments.

  • 在測試回合之間,我們一次只修改一個變數,以便容易比較測試回合之間的結果。Between test runs, we modified only one variable at a time to make it easy to compare results between test runs.

  • 資料庫伺服器並非叢集的一部分,因為就這些測試的目的而言,冗餘不是必要的。The database servers were not part of a cluster because redundancy was not necessary for the purposes of these tests.

  • 在測試期間未執行搜尋編目。當然,它可能會在生產環境中執行。若要事項此,我們會降低 SQL Server CPU 使用率中的 '綠色區域' 與 '紅色區域' 以容納搜尋編目通常會在測試期間使用的資源我們定義。Search crawl was not running during the tests. Of course, it might be running in a production environment. To take this into account, we lowered the SQL Server CPU utilization in our definitions of 'Green Zone' and 'Red Zone' to accommodate the resources that a search crawl would usually consume during testing.

規格Specifications

本節提供我們測試實驗室環境中硬體、軟體、拓撲及設定的詳細資訊。This section provides details about the hardware, software, topology, and configuration of our test lab environment.

硬體Hardware

以下章節說明我們的測試實驗室環境所使用的硬體。The following sections describe the hardware that we used in our test lab environment.

重要

我們使用 HYPER-V 主機要虛擬化我們的測試實驗室中的所有網頁伺服器和應用程式伺服器。資料庫伺服器已不在虛擬化。實體主機硬體和虛擬機器虛擬硬體說明分開這一節。We used Hyper-V hosts to virtualize all web servers and application servers in our test lab. Database servers were not virtualized. The physical host hardware and virtual machine virtual hardware are described separately in this section.

Hyper-V 主機Hyper-V Hosts

我們為了測試使用六個同名設定的 HYPER-V 主機。每個主機執行到兩個虛擬機器。We used six identically configured Hyper-V hosts for our tests. Each host runs one to two virtual machines.

主機硬體 * * **Host Hardware* Value
處理器Processors
2 個四核心 2.49 GHz 處理器2 Quad-core 2.49 GHz processors
RAMRAM
32 GB32 GB
作業系統Operating System
Windows Server 2008 R2 SP1Windows Server 2008 R2 SP1
網路介面卡數目Number of network adapters
22
網路介面卡速度Network adapter speed
1 Gigabit1 Gigabit

虛擬網頁伺服器與應用程式伺服器Virtual web servers and application servers

我們的測試伺服器陣列使用 8 個虛擬網頁伺服器。我們也新增一部執行分散式快取服務的專用伺服器。Our test farm uses eight virtual web servers. We also use a dedicated virtual server that runs the Distributed Cache Service.

注意

在實際執行環境中,執行分散式快取服務的專用伺服器通常會以高可用性組態部署。在測試實驗室環境中,我們使用單一專用伺服器來執行分散式快取,因為高可用性並非重要因素。Production environments typically deploy dedicated servers that run the Distributed Cache Service in a highly available configuration. In our test lab environment we use a single dedicated server for Distributed Cache because high availability is not an important factor.

VM 硬體VM Hardware WFE1-8 與 DC1WFE1-8 and DC1
處理器Processors
4 個虛擬處理器4 virtual processors
RAMRAM
12 GB12 GB
作業系統Operating system
Windows Server 2008 R2 SP1Windows Server 2008 R2 SP1
SharePoint 磁碟機的大小Size of the SharePoint drive
建立在內容資料庫中且可供轉譯文件的空間。根據預設,可供轉譯文件的快取為 100 GB。我們不建議您增加可用的快取。100 GB
網路介面卡數目Number of network adapters
22
網路介面卡速度Network adapter speed
10 Gigabit (主機間流量受限於主機網路介面卡速度)10 Gigabit (inter-host traffic limited to host network adapter speed)
驗證Authentication
Windows NTLMWindows NTLM
負載平衡器類型Load balancer type
F5 Big IPF5 Big IP
在本機上執行的服務Services running locally
WFE 1-8: Basic 同盟服務。這包括下列: SharePoint 計時器服務、 追蹤服務、 Word Automation Services、 Excel Services 及 Microsoft SharePoint Foundation 沙箱化程式碼服務。WFE 1-8: Basic Federated Services. This includes the following: SharePoint Timer Service, Trace Service, Word Automation Services, Excel Services, and Microsoft SharePoint Foundation Sandboxed Code Service.
DC1︰分散式快取服務。DC1: Distributed Cache Service.

資料庫伺服器Database servers

在我們的測試我們會使用一部實體資料庫伺服器並執行儲存的 SharePoint 資料庫的預設 SQL Server 執行個體。我們不要追蹤本文中的記錄資料庫。In our tests we use one physical database server and ran the default SQL Server instance that stores the SharePoint databases. We do not track the logging database in this article.

注意

如果您啟用流量報告,建議您將記錄資料庫儲存在個別的邏輯單位編號 (LUN)。大型部署及部分中型部署可能需要專用的記錄資料庫伺服器,以因應大量記錄事件產生的處理器需求。If you enable usage reporting, we recommend that you store the logging database on a separate Logical Unit Number (LUN). Large deployments and some medium deployments might require a dedicated logging database server to accommodate the demand on the processor that a high volume of logging events generates.

在我們的實驗室環境中我們限制了記錄並儲存在記錄資料庫中的個別的 SQL Server 執行個體。In our lab environment we constrained the logging and stored the logging database in a separate SQL Server instance.

資料庫伺服器-預設執行個體Database Server - Default Instance SQL ServerSQL Server
處理器Processors
4 個四核心 2.4 GHz 處理器4 Quad-core 2.4 GHz processors
RAMRAM
32 GB32 GB
作業系統Operating system
Windows Server 2008 R2 SP1Windows Server 2008 R2 SP1
儲存體與幾何Storage and geometry
直接連接儲存裝置 (DAS)Direct Attached Storage (DAS)
1 x 系統磁碟區 (RAID0,1 個主軸,300 GB)1 x system volume (RAID0, 1 spindle, 300 GB)
2 x 內容資料磁碟區 (RAID0,4 個主軸,每個 450 GB)2 x content data volumes (RAID0, 4 spindles, 450 GB each)
2 x 內容記錄磁碟區 (RAID0,2 個主軸,每個 450 GB)2 x content log volumes (RAID0, 2 spindles, 450 GB each)
1 x 暫時資料磁碟區 (RAID0,2 個主軸,每個 300 GB)1 x temp data volume (RAID0, 2 spindles, 300 GB each)
1 x 暫時記錄磁碟區 (RAID0,2 個主軸,每個 300 GB)1 x temp log volume (RAID0, 2 spindles, 300 GB each)
網路介面卡數目Number of network adapters
11
網路介面卡速度Network adapter speed
1 Gigabit1 Gigabit
驗證Authentication
Windows NTLMWindows NTLM
軟體版本Software version
SQL Server 2008 R2SQL Server 2008 R2

拓撲Topology

下圖顯示我們的測試實驗室環境的拓撲。The following diagram shows the topology in our test lab environment.

測試實驗室拓撲有 4 個 Hyper-V VM (每個各裝載 2 個網頁伺服器),並還有 1 個當成網域控制站的 VM。實體 DB 伺服器執行 SQL Server 2008 R2 SP1 (1 個系統磁碟區、2 個內容資料磁碟區、2 個內容記錄磁碟區、1 個暫時資料磁碟區、1 個暫時記錄磁碟區)

設定Configuration

下表顯示我們對我們實驗室環境中的資料庫伺服器的重大的組態變更。這些設定變更允許以取得最佳的測試效能並清除測試參數與結果之間的關係。請注意 MAXDOP 設定必要的 SharePoint Server 2013。其他設定變更只會套用到我們的測試實驗室環境和可能不會影響實際執行環境。The following table shows the significant configuration changes that we made to the database server in our lab environment. These configuration changes allow for optimal test performance and clear relationships between test parameters and results. Note that the MAXDOP setting is required for SharePoint Server 2013. The other setting changes only apply to our test lab environment and may not affect your production environment.

設定Setting Value 附註Notes
網站集合Site collection
179 (環境中的總數)179 (total in environment)
我們測試環境中的網站集合使用預設設定與 Windows 宣告型驗證。The site collections in our test environment use default settings and Windows claims authentication.
BLOB 快取BLOB caching
開啟On
預設為關閉。若您啟用 BLOB 快取,瀏覽器呼叫資料庫伺服器以取得可能經常要求之靜態頁面資源的情況就會減少,進而改善伺服器效率。The default is Off. If you enable BLOB caching, you improve server efficiency by reducing calls to the database server for static page resources that browsers may request frequently.
平行處理原則 (MAXDOP) 的最大程度Max degree of parallelism (MAXDOP)
11
這個參數會設定 SQL Server 執行個體或包含 SharePoint Server 2013 內容資料庫執行個體上。預設值為 0,可讓 SQL Server,以決定平行處理原則的最大程度。SharePoint Server 2013 需要 MAXDOP 設為 1 包含 SharePoint Server 2013 資料庫的 SQL Server 執行個體。This parameter is set on the SQL Server instance or instances that contain SharePoint Server 2013 content databases. The default value is 0, which enables SQL Server to determine the maximum degree of parallelism. SharePoint Server 2013 requires MAXDOP to be set to 1 for SQL Server instances that contain SharePoint Server 2013 databases.
如需如何設定 MAXDOP 設定的 SQL Server 2008 R2 的詳細資訊,請參閱 <平行處理原則選項的最大程度For more information about how to configure the MAXDOP setting for SQL Server 2008 R2, see max degree of parallelism Option.
如需如何設定 MAXDOP 設定的 SQL Server 2012,請參閱 < Configure parallelism Server Configuration Option 的最大程度For information about how to configure the MAXDOP setting for SQL Server 2012, see Configure the max degree of parallelism Server Configuration Option.

工作量Workload

本節說明我們執行針對 SharePoint Server 2013 的實驗室測試。測試詳細資料是一般的部門共同作業環境。This section explains the lab tests that we ran against SharePoint Server 2013. The test details are typical of a divisional collaboration environment.

實驗室針對 SharePoint Server 2013 的部門共同作業執行了測試。測試詳細資料顯示伺服器在九種案例中收到的要求。

資料集Dataset

我們在測試實驗室環境中使用的資料集充分代表一般的部門共同作業環境。此資料集包含各個網站集合、網站、清單、程式庫、檔案類型及大小。The dataset that we used for our test lab environment represents a typical divisional collaboration environment. This dataset contains various site collections, sites, lists, libraries, file types, and file sizes.

資料集特性Dataset Characteristics Value
資料庫大小 (合計)Database size (combined)
174 GB174 GB
MDF 大小MDF size
154 GB154 GB
LDF 大小LDF size
20 GB20 GB
BLOB 大小BLOB size
152 GB152 GB
內容資料庫數目Number of content databases
22
網站集合數目Number of site collections
179179
Web 應用程式數目Number of web applications
11
網站數目Number of sites
1,4711,471

結果與分析Results and analysis

以下結果被按照概觀> 一節中說明的延展方法。The following results are ordered based on the scaling approach that is described in the Overview section.

網頁伺服器向外延展Web server scale out

下列章節說明當我們向外延展測試實驗室中的環境網頁伺服器數目時,所獲得的測試結果。The following sections describe the test results which we obtained when we scaled out the number of web servers in our test lab environment.

測試方法Test methodology

  • 新增使用相同硬體規格的網頁伺服器,然後在不變更伺服器陣列或測試參數的情況下,再次執行測試。Add web server that use the same hardware specifications, and run the test again without changes to the farm or test parameters.

  • 測量測試伺服器陣列中每部伺服器的 RPS、延遲及資源使用率。Measure the RPS, latency, and resource utilization on each server in the test farm.

分析Analysis

我們在測試中發現下列結果︰In our tests we found the following:

  • 環境延展到每部資料庫伺服器有十部網頁伺服器。輸送量的增加相當地線性。The environment scaled to ten web servers per database server. The increase in throughput was relatively linear.

  • 即使增加到最大測試規模的十部網頁伺服器,新增更多的資料庫伺服器並未增加傳送量。瓶頸一般是限制在網頁伺服器資源。Even up to the maximum tested scale of ten web servers, the addition of more database servers did not increase throughput. The bottleneck was generally confined to web server resources.

  • 在整個測試中,綠色區域中的平均延遲幾乎是固定的。網頁伺服器的數量與傳送量並未影響綠色區域的延遲。紅色區域的延遲資料則顯示了預期的趨勢線。在單一網頁伺服器下,延遲非常高。介於 2 部到 8 部網頁伺服器之間的曲線則輕鬆保持在紅色區域準則之內。The average latency at Green Zone was almost constant throughout the whole test. The number of web servers and throughput did not affect Green Zone latency. Red Zone latency data shows an expected trend line. Latency is very high at a single web server. A curve between 2 and 8 web servers remains comfortably within Red Zone criteria.

    注意

    當您將分散式快取服務從伺服器陣列的網頁伺服器移至分散式快取專用伺服器延遲可能會稍微影響。因為已先前內部的每部網頁伺服器、 的分散式快取流量開始跨網路可能會發生此情況。若要判斷此取捨是否重要環境中測試向外延展效能。請注意我們的測試環境的延遲增加稍微時的分散式快取服務已移轉至專用伺服器。降低每個新增的網頁伺服器與名義新增的延遲已位移來降低饋送的處理和記憶體負載網頁伺服器上的延遲。> 如需分散式快取容量規劃的詳細資訊,請參閱 <規劃摘要及 SharePoint Server 中的分散式快取服務Latency may be slightly affected when you move the Distributed Cache service from a farm's web servers to a server that is dedicated to the Distributed Cache. This may occur because Distributed Cache traffic, which was previously internal to each web server, begins to cross the network. Test scale-out performance in your own environment to determine whether this tradeoff is significant. Note that latency in our test environment increased slightly when the Distributed Cache service was migrated to a dedicated server. Latency decreased with each added web server as the nominal added latency was offset by the decreased processing and memory load on the web servers. > For more information about Distributed Cache capacity planning, see Plan for feeds and the Distributed Cache service in SharePoint Server.

  • 在 SharePoint Server 2013 中的快取及資料庫使用狀況特性的改良功能,因為資料庫伺服器圖層上的平均負載很低。我們找到在測試期間它不是必要向外延展資料庫伺服器。Because of improvements in caching and database usage characteristics in SharePoint Server 2013, the average load on the database server layer is low. We found that during our tests it was not necessary to scale out the database servers.

  • 新增虛擬網頁伺服器是否能增加效能,部分取決於主機硬體資源,以及在同一主機上執行之其他虛擬電腦的資源使用狀況。虛擬伺服器需要針對虛擬化進行額外的計劃與管理策略。Performance gains when you add virtual web servers depend partly on host hardware resources and on the resource usage of other virtual computers that are running on the same host. Virtual servers require additional planning and management strategies that are specific to virtualization.

    如需 HYPER-V 效能和容量規劃的詳細資訊,請參閱SharePoint 2013 的 HYPER-V 虛擬化需求使用最佳作法設定 SharePoint 2013 虛擬機器與 HYPER-V 環境For more information about Hyper-V performance and capacity planning, see Hyper-V virtualization requirements for SharePoint 2013 and Use best practice configurations for the SharePoint 2013 virtual machines and Hyper-V environment.

注意

本節中的結論專屬於環境所組成的硬體。環境可能會有可達到相同的輸送量如果環境使用多個但小於強大的 HYPER-V 主機伺服器或較少,但更強大的 HYPER-V 主機伺服器。增加資料庫伺服器上的硬體資源有用大幅影響結果。Conclusions in this section are specific to the hardware that makes up the environment. The environment might have achieved the same throughput if the environment used more but less powerful Hyper-V host servers, or fewer but more powerful Hyper-V host servers. An increase of hardware resources on the database server wouldn't greatly affect the results.

結果、圖形與圖表Results, graphs and charts

下圖中,x 軸顯示伺服器陣列中網頁伺服器數量的變化。規模從一部虛擬網頁伺服器與一部實體資料庫伺服器 (1x1) 開始。最大規模為八部虛擬網頁伺服器、一部專用虛擬分散式快取伺服器 (新增於四部網頁伺服器時) 以及一部實體資料庫伺服器 (8x1x1)。The x axis of the following graphs show the change in the number of web servers in the farm. The scale starts with one virtual web server and one physical database server (1x1). The maximum is eight virtual web servers, one dedicated virtual Distributed Cache server (added at four web servers) and one physical database server (8x1x1).

注意

本節圖表中的代表每個資料點的平均值的測試持續時間。所有圖形都包含顯示 RPS 和延遲、 伺服器資源使用率和 SQL Server 磁碟使用量等因素之間的關係的綠色與紅色區域的 RPS 基準。The graphs in this section represent the average values for each data point over the duration of the test. All graphs include the RPS baseline for both Green and Red zones to show the relationship between RPS and factors such as latency, server resource utilization, and SQL Server disk usage.

1.RPS1. RPS

下圖顯示向外延展對 RPS 基準線的影響。The following graph shows how scaling out affects the RPS baseline.

插圖顯示隨著前端網頁伺服器及網域控制站向外延展,每秒要求數也增加

2.延遲2. Latency

下圖顯示向外延展對延遲的影響。請注意,綠色區域延遲幾乎一直很平穩,而紅色區域延遲則在可接受的限制內增加。The following graph shows how scaling out affects latency. Note that Green Zone latency remains almost flat, but Red Zone latency shows increases that are within acceptable limits.

向外延展前端網頁伺服器及網域控制站會影響延遲。綠色區域保持平穩,紅色區域則顯示波動。

3.網頁伺服器處理器和記憶體使用率3. Web server processor and memory utilization

下圖顯示向外延展對網頁伺服器上平均處理器與記憶體使用率的影響。請注意,雖然 RPS 增加,但綠色區域的處理器使用率和平均記憶體使用率仍保持相當固定。The following graph shows how scaling out affects average processor and memory utilization on the web servers. Note that Green Zone processor utilization and average memory utilization remains relatively constant as RPS increases.

紅色區域的處理器使用率趨勢是向下。這個向下趨勢反映出,最大負載下網頁伺服器處理器的平均需求是隨著伺服器數量增加而逐漸降低。The trend of processor utilization in the Red Zone is down. This downward trend reflects the fact that the average demand of the web server's processor at maximum load gradually declines as the number of servers increases.

圖形顯示將前端網頁伺服器向外延伸對處理器與記憶體使用量的影響。隨著每秒要求數及記憶體使用量增加,綠色區域還是保持固定。新增伺服器時,因為網頁伺服器處理器的負荷減少,紅色區域顯示減少。

4.SQL Server I/O 作業每秒 (IOPs) 和處理器使用率4. SQL Server I/O operations per second (IOPs) and processor utilization

下圖顯示平均磁碟 IOP (總計與讀寫) 和處理器使用率值,如何隨著網頁伺服器數量向外延展而變化。我們使用下列效能計數器來測量 IOP 值︰The following graphs show how average disk IOPs (both total and reads/writes) and the values of processor utilization change as the number of web servers is scaled out. We use the following performance counters to measure IOPs values:

  • 實體磁碟︰每秒磁碟讀取數PhysicalDisk: Disk Reads / sec

  • 實體磁碟︰每秒磁碟寫入數PhysicalDisk: Disk Writes / sec

測試持續時間內每個計數器的值進行平均,然後加總得到 IOPs 總計。The values of each counter over the duration of the test are averaged and then added together to produce total IOPs.

注意

因為 SQL Server 記憶體使用率的資料可用測試次,此資料不是包含在此圖中。Because the data for SQL Server memory utilization wasn't available at the time of our tests, this data is not included in this graph.

重要

這些 IOPs 測試結果並不能代表實際執行環境,因為我們的資料集比實際執行伺服器陣列的資料集要小得多。這使得在網頁伺服器上可以比在實際執行環境中快取更大百分比的資料。因為我們將更大百分比的資料快取在網頁伺服器上,所以本節中的 IOPs 結果是根據可用的測試資料所計算出來的平均值。我們預期我們的 IOPs 結果一般會比實際執行環境下的 IOPs 來得低。在試驗環境下完整測試您自己的伺服器陣列可能會得到不同的結果。These results for IOPs tests are not representative of a production environment because our dataset was much smaller than that of a production farm. This made it possible for a larger percentage of the data to be cached at the web servers than would be possible in a production environment. Because we cached a larger percentage of the data at the web server, the IOPs results in this section are calculated averages that are based on available test data. We expect that our IOPs results are generally lower than IOPs in a production environment. Thorough testing of your own farm in a pilot environment may produce different results.

請注意,在本節的圖表中,IOPs 與資料庫伺服器處理器使用率在前端網頁伺服器達到 6 部時,都顯示下降,而 RPS 卻持續增加。上圖的網頁伺服器處理器使用率也反映這個差異。Note that in the graphs in this section, both IOPs and database server processor utilization show a drop at six front-end web servers, while RPS continues to increase. This variation is also reflected in web server processor utilization as shown in the previous graph.

這顯示出伺服器陣列的規模已達到某個點,也就是就算只使用基準負載與資料集,也對陣列伺服器資源造成最大壓力。伺服器資源必須要有較低的平均使用率,才能支援伺服器陣列的負載。This shows that the scale of the farm has reached a point at which maximum pressure on the farm server resources was achieved by using the baseline load and dataset. A lower average utilization of server resources is required to support the load on the farm.

從此趨勢來看,可推導出下列結果︰It is possible to conclude the following from this trend:

  • 如果我們在第六部網頁伺服器的規模點增加測試負載,將可達到更大的 RPS,同時維持伺服器資源使用率的平穩曲線。Had we increased the test load at the sixth web server scale point, greater RPS could have been achieved while maintaining a flat curve in the utilization of server resources.

  • 如果我們繼續向外延展網頁伺服器數量,同時維持相同的測試負載,RPS 就會繼續增加,而對伺服器資源的壓力會繼續呈向下的趨勢。Had we scaled out the number of web servers further while maintaining the same test load, RPS would have continued to increase while pressure on server resources would have continued a downward trend.

  1. SQL Server IopSQL Server Total IOPs

    下圖顯示向外延展對 IOPs 總計的影響。The following graph shows how scaling out affects total IOPs.

    圖形顯示綠色與紅色區域的 SQL Server IOPs 總數。兩個區域都維持增加一直到 4 個前端網頁伺服器處,然後持平,再於 8 個網頁伺服器處逐漸減少。

  2. SQL Server IOPs 分為讀取與寫入作業SQL Server IOPs broken down into read and write operations

    下圖顯示向外延展對 IOPs 在每秒讀取數與每秒寫入數的影響。The following graph shows how scaling out affects IOPs for reads per second and writes per second.

    圖形顯示將前端網頁伺服器向外延伸,將如何影響跟每秒讀取與寫入數有關的 IOPs。每秒讀取與寫入數會維持往上到 4 個前端網頁伺服器處,然後每秒讀取數會逐漸減少,而每秒寫入數會繼續增加。

  3. SQL Server 處理器使用率SQL Server processor utilization

    下圖顯示向外延展影響 SQL Server 處理器使用率。The following graph shows how scaling out affects SQL Server processor utilization.

    插圖顯示隨著新增越多網頁伺服器,SQL 處理器及每秒讀取次數的趨勢也會往上

另請參閱See also

概念Concepts

規劃 SharePoint Server 2013 中規劃效能Performance planning in SharePoint Server 2013

效能及容量測試結果與建議 (SharePoint Server 2013)Performance and capacity test results and recommendations (SharePoint Server 2013)

評估企業內部網路共同作業環境 (SharePoint Server 2013) 的效能和容量需求Estimate performance and capacity requirements for enterprise intranet collaboration environments (SharePoint Server 2013)