Hardware and network configuration

ABBYY Recognition Server can use the Named Pipes or TCP/IP protocols for connections among its components.

By default, the Named Pipes protocol is used via port 445.

It is possible to change the protocol to TCP/IP. In this case the ports are chosen by the administrator and are defined in the Windows registry.

Detailed instruction on how to change your connection protocol can be found in the Connection Settings article of the ABBYY Recognition Server Help file.

Yes, ABBYY Recognition Server can run on a standalone computer without a network connection. In this case all the required program components must be installed on the same PC.

Recognition Server architecture makes it possible to scale the system and increase the processing power by connecting additional Processing Stations to the Server Manager, or by increasing the number of CPU cores on the Processing Stations. The Server Manager distributes the workload among all available Processing Stations and their CPU cores evenly, so that the maximum productivity is achieved.

The number of CPU cores and Processing Stations that can be used simultaneously is regulated by your license. If the license does not limit these parameters, you can connect an unlimited number of stations and enable all CPUs on the stations for OCR.

Technically, there is no known limit to how many Stations can be used by one Server Manager. Successful tests have been conducted with over 100 Processing Stations connected at the same time. The processing speed increase is almost proportional to the number of CPU cores enabled on all the Processing Stations altogether. That means, for example, that connecting 10 single-core stations would allow you to process almost 10 times as many files as one station can process in the same time.

The multi-core, multi-CPU and HyperThreading technologies are fully supported by ABBYY Recognition Server. For example, if you install a Processing Station on a server with two quad-core CPUs and your license allows at least 8 cores to be used, Recognition Server will load all 8 cores to the full extent. If you wish to leave one or two cores for tasks other than OCR, you can do so by unselecting the corresponding checkboxes in the Processing Station Properties dialog box.

Please note that a multi-core CPU usually provides slightly lower productivity increase than an equal number of single-cores. For example, productivity of a quad-core is usually less than 4 x (productivity of each core).

Note: A single-core CPU with HyperThreading will be shown as one core in the Processing Station Properties dialog box. However, Recognition Server will take full advantage of the additional power provided by the HyperThreading technology.

Recognition Server uses the same algorithm of job distribution among CPU cores, whether these CPU cores are located on one Processing Station or on different Stations. Therefore, it does not matter for Recognition Sever whether a multi-core station or several single-core stations are used. But when deciding between a multi-core PC and single- or dual-core PCs, consider the following:

• A quad- or eight-core server is usually more expensive than the equal number of single- or dual-core PCs;

• A multi-core CPU usually provides a slightly lower productivity increase than the equal number of single-cores. For example, productivity of a quad-core is usually less than 4x(productivity of each core).

• Should a multi-core computer experience a hardware failure, it becomes completely inoperable and its processing power becomes unavailable until the computer is repaired. Several computers with fewer cores each are much less likely to fail all at the same time.

Therefore, using several single- or dual-core Processing Stations is usually more efficient than using one multi-core station.

If your license permits, you can add several Processing Stations to Recognition Server.

When image files are submitted to ABBYY Recognition Server, the Server Manger queues them for processing. The first file in the queue will be sent to the first available Processing Station for processing. If there are several Processing Stations in the system, the Server Manager will evenly distribute the images from the queue among these Processing Stations.

A Processing Station can run several OCR processes (the number of processes can be adjusted in the Processing Station properties dialog box of the Remote Administration Console).

For optimal performance, the recommended number of processes for a station is N+1, where N is the number of CPU cores enabled for OCR on the station (e.g. a dual-core Processing Station typically runs 3 OCR processes).

Note: For HyperThreading CPUs the recommended number of processes in 2N+1, where N is the number of physical CPU cores.

Usually each process on a Station gets from the queue one file at a time. It does not matter whether the file is single- or multi-page. For example, suppose 10 dual-core Processing Stations are connected to the Server Manager. Each Station runs 3 processes, so each will receive 3 files from the queue and recognize them in parallel. The total number of processes will be 10x3=30, i.e. Recognition Server will process 30 files concurrently.

However, if the file is fairly large and there are no more than 5 jobs waiting in the queue, the file will be split into chunks, and the chunks will be sent to different processes, in order to get the work done faster. By default, the size of a chunk is 25 pages. Therefore, for example, a 100-page file will be split in 4 chunks and sent to four processes. The processes to which the file chunks are sent are chosen on the "first available" principle, so they may be running on one or on different Processing Stations.

For optimal performance it is recommended that you install the Server Manager and the Processing Station on different computers in the LAN. Then the Processing Station will be able to use 100% of the CPU time for OCR providing maximum productivity, while not struggling for resources with the Server Manager.

However, if you have a limited number of computers at hand, or if you have one multi-core server that you want to use for a standalone installation, then you can install the Server Manager and the Processing Station on the same computer. In this case you should check the option “Low processing priority” in the Processing Station Properties dialog box. That will make the OCR process run with lower priority relatively to other system processes, and Windows will allocate enough CPU time for the Server Manager to do efficient load balancing.

The processing speed on a single CPU depends mostly on the following two factors:

1. CPU clock speed: the higher the clock speed, the faster the OCR.

2. Number of CPU cores. Tests have shown that running a Processing Station on a dual-core CPU results in about 1.85 times faster processing as compared to a single core CPU.

Minimum hardware requirements for the Server Manager:

• CPU with clock speed of 500 MHz or higher

• 128 MB RAM

• 115 MB HDD space for installation and operation

The above requirements are sufficient for a standalone installation of Recognition Server, e.g. for program evaluation or for processing medium volumes of images. However, if Recognition Server is scaled up for processing large piles of images and working with many Processing Stations, more system resources will be required. Follow the guidelines below:

• CPU of the Server Manager PC should be powerful enough to do job distribution amongst many stations. An average server CPU like Intel Xeon 3 GHz should be sufficient.

• The speed of disk I/O operations on the Server Manager PC should allow moving lots of files between folders quickly. When performing backlog conversion of large volumes of images, consider creating a RAID array.

• Free disk space on the Server Manager PC should be at least equal to the total size of all images put in the Input folder, because all the images will be copied from the Input folder to the Server Manager temporary folder.

As Recognition Server is designed for unattended operation, it has a number of fault tolerance features that ensure robustness of the system:

• The Server Manger and Processing Stations work as Windows services, and can start automatically at Windows startup. For example, in the case of a temporary power loss, the above mentioned components will get up and running automatically after the PC is rebooted;

• The Server Manager periodically checks connection with the Processing Stations to make sure that they are running and are available. If a Processing Station is temporarily down for some reason, the Server Manager will re-establish connection after it is back to work. No administrator’s involvement is required to re-connect a Processing Station.

Recognition Server processes documents in a safe way, so that no data is lost in the case of a system failure:

• All image files from the input source are imported into the Server Manager temporary folder and safely stored there during the whole conversion process. If the PC on which the Server Manger is installed experiences a failure, the files that were being processed will not be lost, they will be preserved in the temporary folder and stay in the queue. The processing of the files will continue after the PC is back to work;

• The Processing Stations and Verification Stations receive copies of images from the Server Manger. This ensures that no files are lost if an error occurs during the recognition or verification process. If a Processing Station or Verification Station experiences a failure while processing an image, the Server Manger will automatically route a copy of the image to another available station. If no stations are available at the moment, the image will remain in the queue and wait for the first available station.

No, ABBYY Recognition Server does not support clustering technologies.