Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.adass.org/adass/proceedings/adass00/reprints/P2-39.pdf Дата изменения: Wed May 30 01:31:46 2001 Дата индексирования: Tue Oct 2 11:23:15 2012 Кодировка: Поисковые слова: р р р с р с р р р с р с р р р п п п п п п п п п п п п п п п п

Astronomical Data Analysis Software and Systems X ASP Conference Series, Vol. 238, 2001 F. R. Harnden Jr., F. A. Primini, and H. E. Payne, eds.

The Chandra Automatic Pro cessing/Archive Interface
Sreelatha Subramanian, David Plummer Harvard-Smithsonian Center for Astrophysics, 60 Garden St. MS-81, Cambridge, MA 02138, Email: latha@head-cfa.harvard.edu Abstract. The Chandra Automatic Data Processing System (AP) requires quick access to previously generated data. Potential inefficiencies are avoided by introducing a layer between the pipelines and the archive. This archive interface layer includes an archive request queue, a data archiving server (darch), and an archive "cache". The design and functional operation of each of these components are presented in this paper.

1.

Cache

The first step in improving the AP system is the implementation of a cache. In theory, the cache operates as a typical cache--data is retrieved from a slowaccess storage facility and is stored to a higher-speed retrieval location, with the expectation that it will be used again. In our implementation, the cache is actually a directory where the data files reside, as shown in Figure 1. While a pipeline is processing, it will store its data products to the cache directory. These data products can then be stored to the archive at any time, separating data product ingestion from processing. When another pipeline requires those data products, the cache is checked first. If the data products are not in the cache, then the pipeline retrieves the files from the archive database. This method is especially significant when processing data for the first time since the system can access all the data products from cache, avoiding the archive database. If re-processing is required, data will be retrieved from the archive if it is no longer in cache. 1.1. Inactive Cache

The implementation of a cache introduces the issue of memory usage. The cache server loads all the files in the cache directory into its memory, which allows for quick search and retrieval of the data. However, as the size of the cache increases, the cache server requires more memory. To reduce the number of files in cache, an inactive cache directory structure is created. The cache server does not load files that exist in the inactive directory into its memory. It stores a few pieces of information about the file along with the filename, in order to facilitate retrieval. The implementation of the inactive cache eliminates the issue of memory usage, while maintaining the usefulness of the cache.

303 c Copyright 2001 Astronomical Society of the Pacific. All rights reserved.


304

Subramanian and Plummer

PIPELINE/ARCHIVE INTERFACE

CACHING SYSTEM

CACHE

INACTIVE FILES

VIRTUAL FILES

ARCHIVE DATABASE

CACHE SERVER

DATA ARCHIVING SERVERS
CACHE DARCH RETRIEVE DARCH ARCHIVE DARCH

FILE QUEUES
CACHE REQUEST RETRIEVE REQUEST ARCHIVE REQUEST FILE QUEUE FILE QUEUE FILE QUEUE

L0 PIPELINES
SENDS NOTIFICATIONS
PIPELINES OFF KICKS

L1 PIPELINES

OBSERVATION STATUS TRACKER

OTHER PIPELINES

Figure 1.

AP/Archive Interface


Chandra Automatic Processing/Archive Interface 1.2. Virtual Cache

305

A significant improvement to the cache system is made through the implementation of virtual files. In our system, virtual files are simply files that do not exist in the cache, but whose filename and metafile still exist. With this information, the cache can be queried to retrieve a file based on time, data product id or other information. The cache returns the filenames of the files it found, and also returns an error code which indicates that the files are virtual. The AP system, recognizing that the retrieved files are virtual, can then use the filename to retrieve the file from the archive database. This modification is significant because retrieving a file from the archive using the filename as the key is more efficient than retrieving using another key (such as data product id, or time). The other important factor is that the implementation of virtual files allows us to search for file(s) by using cache specific tag information. Tag information, which is stored in the metafile, is not entered into the archive. Therefore it was previously not possible to retrieve files from the archive using tag information. With virtual files, however, this becomes possible. A search by tag function searches the cache, using the tag. As this information is in the metafile, the function will return a corresponding filename, which can then be used to retrieve the file from the archive. 2. File Queues

Another important modification to the AP System is the implementation of file queues. File queues are implemented to eliminate the time a pipeline must wait for its data products to be ingested into the archive database. Instead of waiting for a file to be ingested, the ingestion request is sent to a file queue. The request is then picked up at a later time by the data archiving server, or darch, and processed. The implementation of file queues is simply a file which contains all the requests. The pipeline will not need to wait for ingestion since it has already stored the file to cache. Therefore, ingestion of the file does not need to occur simultaneously, and the file queue allows for another process to handle the requests at any time. 3. Conclusion

The archive interface layer was not part of the original design. However, the necessity for it was identified during the processing of ground calibration data of the system before launch. This allowed us to implement and integrate the archive interface layer into the AP system, to avoid possible bottlenecks during the mission. 4. Acknowledgments

This pro ject is supported by the Chandra X-ray Center under NASA contract NAS8-39073.