GIS transduction Performance Tuning

Status and Objectives

DellR920 server, virtualized VMware VSPHERE 5.5 system, guest system Windows Server R2. Deploy software ArcGIS 10. Image transduction test: The original cut graph takes 70 hours and now takes 17 hours. There is still room for further optimization. Debug target:

1. Verify the optimal graph parameter configuration under the virtualized environment and issue the best performance configuration method for ArcGIS10 under the virtualized environment.

2. Verify the effectiveness of SSD applications, test SSD SRT technology, transduction cache technology, ESXi host cache technology, the optimal performance ratio of SAS 12GB specification mode, and conclude the optimal SSD configuration optimization method in transduction and R920 server environment.

3. Verify the best method of optimization for your ArcGIS system. Verify best practices for optimized debugging in ArcGIS.

4. Shorten the optimization time again and try to reach the shortest time by different technical means.

Tuning content A. Virtualization of

1. The VSPHERE server version is selected as VSPHERE5.5 Updata 2 with VMtools.

2. When the host is out of physical memory, use the ESXi host cache technology.

3. Hardware optimization

4. Virtual machine configuration

5. Vmfs file system optimization, enable SAS and RAID card Alwayswrite back write buffering function. Because transduction is a continuous write operation, the operation is a required item.

B. Operating system Windows R2

1. Modify the registry to change the performance of the TCP network protocol stack.

2. Enable the max value of default CPU performance for Windows 2012. Windows 2012 Power-saving considerations will turn off some CPU cores.

3. Optimize page file, recommended to set to Minimum = physical memory *2+50m, max = physical Memory *4+50m, can refer to MS Official document: http://support.microsoft.com/kb/2860880

4. Io Optimizer, the registry modifies the number of default IO requests for Windows2012.

5. NTFS file system optimizations

C. Solid-State Drive optimization

SSD-based optimization solves the problem of write amplification generated by erase-before-write, with different types of IO separation, reducing the performance impact of write operations.

1. Modify the sequential logging to in-page logging to avoid repeated erasure of the same position.

2. A large number of In-place update random writes are combined into a few sequential writes in a cache-write manner.

3. The use of high-capacity SSD random read and write capabilities, reduce write increase read, so as to achieve overall performance improvement.

4. Control Panel-management tools-Services, disable SuperFetch, Disk Defragmenter, and Windows Search services: For SSDs due to changes in storage mode, the original disk fragment cleanup becomes redundant, even harmful, So we need to disable disk fragmentation cleanup, and the SuperFetch pre-read and Windows Search cables for traditional mechanical hard drives are also now unnecessary to disable.

5. Use faster read and write memory as the SSD cache. The specific principle is that when the system reads and writes the SSD, the data is stored in a virtual cache of memory. Some of the commonly used data can be read and written repeatedly in memory, thus reducing the number of reads and writes to the SSD. The advantage of this is that it is faster to read and write, and the second is to prolong the life of the SSD.

D. ArcGIS Optimization

1. Maplex Premium Note. Map documents that are edited using the Maplex Advanced Annotation feature may result in a lower efficiency of the cut graph. To determine whether the transduction is the result of Maplex slow, the simplest way is to close the Maplex engine after the cut graph, and the unit time to generate a slice file size and the original to do a comparison, if the speed difference between the You can then try to remove the map document from a layer other than the current cut scale, and then grade the graph.

2. Format of the published file. Most of the time we do the graph, are all on the server, quoting ESRI after-sales technical engineer, the service on the multi-core server via MXD may be an exception, Esri admits that it is a bug, the recommended use of MSD-based map services. So do the slices on the server, be sure to save the map document as an MSD format, and then do the cut.

3. Set the maximum number of instances. In general, we will say that the number of instances of the Transduction service is set to the number of CPU cores +1, because one SOC process is dedicated to recycling and does not participate in the actual work, so this setting can play the maximum performance of the tile server. But in the actual course of operation, we found that this way on some types of servers will cause the SOC process to be created and recycled, and eventually only one process in the transduction situation. At this point we can only find a suitable value by constantly trying to set the number of different instances.

4. How to cut the picture. As I know, there are two ways to do a graph, either by performing an update slice directly in the map service properties or by using the server tools provided in the Arctoolbox tool. In principle, these two methods are called the same GP process, there is no difference, but in actual use, we update the slice through the service properties often encounter the ArcMap program closed unexpectedly, so it is recommended to use the tool to cut the diagram.

5. Compact & Loose. ArcGIS10 introduced a new cache format, the original independent cache image, saved to bundle format binary files, each bundle file storage 128*128 cache pictures. The benefits of this are easy cache management, reduced disk space usage, improved cache read efficiency, and more, compared to thousands of tile files. So we should also use a compact format to create slices during the drawing process.

6. Picture format. Different picture formats, directly affect the size of the cache file, in general, the system is used by default PNG32 to save tile files, but if we want to reduce some space, you can consider using JPG or PNG8, PNG24 and other formats. Using JPG Although the compression ratio is the largest, but the created tile file noise is serious, so it is not recommended. Through several attempts, the PNG8 format can not only greatly compress the size of the cache file (probably the PNG32 format of 60%), and the effect of the slice is not small, so is the PNG32 format is a good alternative.

Next work

1. Replace the SAS12GB specification interface for EMLC particle testing.

upgrade firmware to the latest version of the master chip firmware. 6 months 3 days updated firmware, version a3ae,a00.

• Nand:24nm EMLC
• Interface:sas 12gb/s
• Form factor:2.5 "SFF x 7mm High (15mm for 1.6TB model)
• Performance
• Sequential Read (sustained): 900mb/s
• Sequential Write (sustained): 400mb/s
• Random Read 4k (IOPS): 120,000
• Random Write 4k (IOPS): 30,000

2. optimize the graph using the above optimized method of ARCGIS.

3. Try a variety of different scenarios and test targets to cut graphs and test their stability and speed.

GIS transduction Performance Tuning