Usage Note 57621: Slow Performance in SAS® Forecast Server might be caused by project settings or input data
 |  |  |  |
Slow performance in SAS Forecast Server might result from a combination of any of the following topics:
- Failure to follow general requirements and best practices for SAS Forecast Server. See SAS KB0036278, "Requirements and best practices for SAS® Forecast Server".
- SAS® Forecast Studio project settings or input data.
- SAS Forecast Server and/or system configuration issues. For tips about slow performance that is related to configuration, see SAS Note 57654: "Slow Performance in SAS® Forecast Server might be caused by configuration settings".
If SAS Forecast Server performs slowly, then here are some areas to investigate:
Logs
- Examine the SAS log and the workspace server log. For information about enabling workspace server logs, see SAS KB0036287, "Enabling SAS® workspace server logs for SAS Forecast Studio".
Inputs
- Having a large number of inputs requires more resources, and can cause performance issues. Do you have too many independent variables or events? Are all these inputs necessary for your analysis? Consider methods for reducing the number of inputs. Use PROC TIMESERIES or PROC SIMILARITY to see the cross-correlations or similarity measures between each X and Y to help you determine which inputs to include in your project. Also, let SAS Forecast Server perform a one-time diagnosis to determine significant inputs. Then, create a new project with just the significant inputs. You can use both methods as a 2-step data reduction technique.
- Since a large number of inputs requires more resources, consider using Exponential Smoothing models at the lowest level of the hierarchy. See the "Models" section below for more information.
- In the historic region, independent variables that do not vary, or vary very little, or contain a lot of missing values, can impact performance.
- Because events are created in memory, use events instead of dummy independent variables.
- If you are concerned about independent variables, and event effects, that occur on a less frequent basis, then see the section "frequency and size of data" below.
Diagnostics
- Enabling outlier detection can significantly negatively impact performance, particularly when working with weekly time series data, because outlier detection is an extremely I/O intensive analysis.
Size and type of data
- Clean up and accumulate data outside of SAS Forecast Server. See Data pre-processing requirements for SAS Forecast Studio SAS Note 56928: "Data pre-processing requirements for SAS® Forecast Studio".
- Large data sets require more time to run. If you can capture the characteristics of your data in a more recent subset of data, then consider using that subset instead of the entire data set.
- Data views might slow down performance. If you have sufficient disk space, then run the data view (possibly overnight) to create a data set. Then, use that data set as input to SAS Forecast Server.
Frequency and size of data
- Perform the forecast at a lower-frequency time interval. Example: instead of forecasting daily, forecast weekly.
- Generate low-frequency data. Leave inputs out of the high-frequency data. Forecast at both frequencies. Then, use temporal reconciliation to synchronize the forecasts. This temporal reconciliation is available only through code (using PROC HPFTEMPRECON).
- Use only a portion of the latest high-frequency data. Forecast at high- and at low- frequencies. Then, use temporal reconciliation to bring the forecasts in synchronization.
- Also, you can use an adjustment method. Create adjustments that are the ratio of high-frequency/low-frequency for a particular time period. Then, apply this adjustment to the low-frequency forecasts to obtain high-frequency forecasts.
- Forecast at low-frequency and profile at high-frequency.
Energy-industry data
- If you have a lot of energy-industry data, then consider using SAS® Energy Forecasting. SAS Energy Forecasting is designed to improve load-forecasting performance for utilities.
Size of project
- If the data set is very large, then consider creating more than one project - typically grouping time series with similar properties. This type of grouping is always a recommended practice for statistical-modeling concerns as well as for reliability.
- In some cases, it might be possible for you to break series into separate series.
- Set up your hierarchy in a way that makes sense for forecasting. Do not build a hierarchy that is based on reporting requirements. See Mistakes in the Forecasting Hierarchy.
- If you run out of memory, then try to reduce the number of BY groups at each node. Example: add an intermediate level in the hierarchy.
Models
- Exponential Smoothing models do not take as long to produce forecasts as do ARIMA and Unobserved Components models. SAS Forecast Studio has a setting that enables you to run only Exponential Smoothing models at the lowest level of the hierarchy when you are using a hierarchy. Often, simple Exponential Smoothing models take very little time to diagnose, and they do a fairly good job for sparse, noisy data.
- Enabling System-Generated Unobserved Components models can significantly negatively impact performance, because this class of model is computationally expensive to estimate.
Overall
- Close all or most of the graphs and plots in the Modeling View (and Series View). Closing these plots might speed up the responsiveness of the user interface.
- Do not select the check box for "Create the component series data set". This data set is very large and requires considerable resources (time and memory) to create.
Libraries
- If you have many libraries, then you might want to unassign the libraries that are not needed. In SAS® Management Console, unassign libraries that point to the Workspace Server that is configured for your SAS Forecast Server environment.
Index files
- SAS Forecast Server procedures do not benefit from indexing the input data. Indexing can actually slow processing of the input data.
- But, creating indexes for the SAS Forecast Studio output data sets can substantially improve some queries. Verify that SAS Forecast Server code is generating index files. Go to the menu option Project ► SAS code, and make sure that "Include code to generate index files" is selected. This option is selected by default.
Operating System and Release Information
| Product Family | Product | System | Product Release | SAS Release | ||
| Reported | Fixed* | Reported | Fixed* | |||
| SAS System | SAS Forecast Server | Solaris for x64 | 12.1 | 9.3 TS1M2 | ||
| Linux for x64 | 12.1 | 9.3 TS1M2 | ||||
| Linux | 12.1 | 9.3 TS1M2 | ||||
| HP-UX IPF | 12.1 | 9.3 TS1M2 | ||||
| 64-bit Enabled Solaris | 12.1 | 9.3 TS1M2 | ||||
| 64-bit Enabled HP-UX | 12.1 | 9.3 TS1M2 | ||||
| 64-bit Enabled AIX | 12.1 | 9.3 TS1M2 | ||||
| Windows Vista for x64 | 12.1 | 9.3 TS1M2 | ||||
| Windows Vista | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Ultimate x64 | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Ultimate 32 bit | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Professional x64 | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Professional 32 bit | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Home Premium x64 | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Home Premium 32 bit | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Enterprise x64 | 12.1 | 9.3 TS1M2 | ||||
| Windows 7 Enterprise 32 bit | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows XP Professional | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2012 Std | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2012 R2 Std | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2012 R2 Datacenter | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2012 Datacenter | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2008 for x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2008 R2 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2008 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2003 for x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2003 Standard Edition | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2003 Enterprise Edition | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows Server 2003 Datacenter Edition | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8.1 Pro x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8.1 Pro 32-bit | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8.1 Enterprise x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8.1 Enterprise 32-bit | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8 Pro x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8 Pro 32-bit | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8 Enterprise x64 | 12.1 | 9.3 TS1M2 | ||||
| Microsoft Windows 8 Enterprise 32-bit | 12.1 | 9.3 TS1M2 | ||||
| Microsoft® Windows® for x64 | 12.1 | 9.3 TS1M2 | ||||
Slow Performance in SAS® Forecast Server might be caused by project settings or input data.
| Type: | Usage Note |
| Priority: |
| Date Modified: | 2016-04-21 10:29:02 |
| Date Created: | 2016-02-10 13:09:59 |