Annual Report
1996

Department of System Integrity
and Networking

November 19, 1996

  • 1 The Department

  • 2 The Report

  • 3 Work plan

    1 The Department

    1.1 Intro

    The department of system integrity and networking of the Institute for High Performance Computing and Databases (formerly known as Institute for Interphase Interactions) is responsible for system hardware and software, data integrity and user support. Along with these (routine for the system department) tasks we perform a wide range of porting and programming activity aimed in making the public access to the supercomputing cluster more comfortable. The networking is one of the most important fields of our activity as well.

    1.2 People

    The Department's staff is probably the youngest one in the Institute. Our employees have graduated the well-known colleges and universities of the Russia, including St. Petersburg State University, Institute for Fine Mechanics and Optics, Baltic University, etc. The high skill of the people may be illustrated by the following list: we have the experience of probably all actual operation systems (SunOS, Solaris, ConvexOS, PARIX, Linux, WinowsNT, etc.), know nearly all the programming languages, including: C, C++, Java, SQL, 4GL, PL1; have the experience of programming under OSF/Motif, IPC, TLI, XDR/RPC, and so on.

    1.3 Structure

    There are three laboratories within the Department: This formal division does not prevent from a tight co-operation between the laboratories. Moreover, the Department may be considered as a unified team. The working and initiative groups on the main directions of our activity normally consist of people that belong to different laboratories.

    1.4 CSA

    The computer center of the Institute is a public-accessible (for educational and academic stuff), federal level computing resource. The latter is named Center for Supercomputing Applications (CSA). Our responsibility is to maintain (and manage) the computing and networking equipment of the center. Following computer systems are presently available: The user access is provided via network (remote access) or directly from the public terminal classes. The Center is connected to the outside world via Russian University Network (RUNNet) that gives us the most direct path to the end users. The networking concept was developed by the Department, and at the present stage includes: FDDI segment, Ethernet (TP, FOIL, BNC) segments, and E1 WAN link to the central networking node of RUNNet. Our networking operation center is built on a base of DEChub 900 from Digital Equipment Corp., and CISCO Systems routers.

    1.5 User support

    The computer systems of the CSA are unique for the Russia and Eastern Europe. That is why a special attention is paid to the detailed studying of the architecture, hardware, and software of the supercomputers. The knowledge obtained is very actual for the users. We provide tutorial materials, deliver lectures, and organize regular workshops devoted to the principles and practice of an application optimization. The online user support includes both system support and application-level support.

    1.6 Directions

    Following are the main directions of our activity:

    2 The Report

    2.1 Projects

    In 1996 the Department contributed to the long-term State programme for the establishing a central federal-level computing and informational resource for the Russian academic and educational network. This accounted the 60\% of the annual budget of the Department. A special financing from the Foundation for Technological Development of the Russian Ministry for Science allowed to purchase and install multiprocessor Parsytec CC/16 system. The development of the networking infrastructure of the center was partially financed by the RFFI grant No. 96-07-89218.

    2.2 New xxxxware: installation and management

    In February 1996 the Convex C 3440 mainframe was delivered. The routine tasks of the system installation and configuration were performed. We changed the configuration of the Convex C 3820 mainframe in order to create a distributed computing cluster uniting the resources of C 3820 and C 3440. The Network Information Service and Network File System service were launched to support the cluster configuration, together with the appropriate re-configuration of CXbatch (a NQS-like batch processing system). In July, the Institute have got a Parsytec CC/16 multiprocessor system. This systems was the first system of MIMD architecture installed in the CSA. Now the computer is available for the local and remote access.

    2.3 Networking

    The attachment of the remote office of the Institute was performed. The network solution was developed basing on NOKIA synchronous modems and CISCO router. This scheme allows to minimize the monthly fee for the link layer (that is less than $50), providing a good throughoutput of 128 Kbaud.

    2.4 User support

    The user support includes the hot lines (both phone and e-mail) for the urgent questions, training in form of lectures and workshops on the most actual problems of programming for vector and parallel computers, and publishing of tutorial papers. Latter are available both in hard copy and in hypertext forms. During the 1996 following new tutorial papers were published: Every day we got a dozens of calls via hot channels for the user support.

    2.5 Programming and porting

    Nowadays, it is not possible to use computers in a static fashion (i.e. to be happy with the pre-installed software). The reason is that the computer world is extremely flexible, offering more and more ideas every day. The things are going faster, when one may accept the result of work of the other. The global networks help people in sharing their efforts. Public software (freeware, shareware) became very popular, and often became a standard one (TeX, gzip, gcc, xpaint, etc.) In order to give the users the common environment (that they used to use on personal computers and workstations), we install the GNU products on our platforms. The Convex systems (as any expensive supercomputer) are not so widely known as a programming platform as, say, SunOS or Linux. That is why most of the publicly available software needs a certain porting efforts for successful compilation and installation. This year following packages were ported to ConvexOS 11.0: In some cases an additional programming was necessary. The problem: there are at least three popular types of cyrillic charsets, that are used to present Russian texts. One is traditional (presently used on UNIX systems and being a standard for mail transfer) KOI8 encoding; two others are used on PCs: so-called IBM-alternative is a de-facto standard for DOS, and cp-1251 charset is used under Windows. This pretty situation cause a good plenty of problems when migrating from PCs to UNIX-based mainframes and back. But the things look even worse for the WWW administrator, because she have to ensure that every client will get documents in a correct (with respect to the client's setup) encoding. Moreover, the information sent back by the client should be converted to the server's encoding as well. The work on this problem resulted in creation of four plug-in cyrillic modules for Apache's hypertext server. This made obsolete our previous (temporal) solution, based on CGI ``master'' script, called every time the client queries for a Russian-language document. The present solution is much more effective (and much faster), because all the conversions are made internally by the server. This software (as far as we know) is the most complete one, that allows flexible configuration of the client recognition, and supports the advanced features like server-side includes, and cyrillic decoder to CGI interface. The Harvest object cache module from Harvest network information retrieval package was ported to ConvexOS and installed to provide the RUNNet users with the powerful proxy server, caching HTTP, FTP, and Gopher requests. Presently, the proxy serves an average of 750 queries per hour. Another programming problem appeared when analyzing the external traffic of the RUNNet. It was found, that 80 percent of the connections are made on WWW ports. Inspite of the existence of proxy server many users ignore the obvious advantages of information caching. There may be different points of view why that is so, but that is so. That is why the pilot project was launched in November 1996 aimed to develop an opaque caching system. As expected, such a system will be able to encapsulate from 30 to 50 percent of the international traffic, giving a reasonable network load reduction. The task of routing a significant fraction of RUNNet's international traffic manifests rather critical demands to the computer platform to be used as opaque cache server. The daily traffic is estimated to be of several Gigabytes with a thousands of simultaneous connections. We plan to use Convex two-processor mainframe for the purpose. The pilot project consists of three parts: The second stage of the project is under development. We plan to finish the pilot project by the December 1, 1996, and to start the experimental runs. The results of these runs will be used to develop the specifications for the further developing of caching software.

    2.6 Databases, Web, FTP, Proxy

    The Department performs the practical work on installation and configuring of the hypertext servers within the CSA, maintenance of the CSA's FTP archive and Proxy server. We provided teaching of users and administrators on how to use proxies, how to get the benefits of our (advanced) hypertext server, gave consultations to the personnel of other departments, performed user training and everyday support. The Institute plan to purchase a professional database server for multiprocessor mainframe in 1997. Now we have installed two servers (as an experiment): Postgress95 and miniSQL in order to: (i) prepare personnel and administrators, and (ii) to solve the most severe local accounting tasks. These two SQL platforms are used for the studying (and development) of the mechanism of the interaction between WWW and SQL servers. These exercises resulted in a development of the WebShell for the miniSQL, being an SQL monitor with CGI-oriented functions for processing of hypertext forms based on POST protocol. Since this interface isolates the actual SQL from CGI, we hope that the most of the underlying modules, and interface forms will be used with professional SQL server as well.

    2.7 Security issues

    The cluster is a public resource, presently containing no private or commercial data. By the present time our main goal was to insure the widest possible access to the CSA. That is why, the cluster was configured as an opened publicly-accessible network. The system security model is presently based on the: (i) proper configuration of the every computer within the CSA, (ii) teaching the users about the current configuration and administrative limits, (iii) proper password aging and typing policy, (iv) regular back-ups. The network space within the CSA is considered to be a ``dirty'' one. The life is not a plain thing. Therefore, sometimes it is necessary to perform a certain system and security works in a very limited terms. Such tasks are not only unpredictable, but cover a wide range of things, from hardware diagnostics to kernel patching. The special ``emergency'' group is formed within the Department, that is capable of solving these problems.

    2.8 International relations

    The cooperation in the field of parallel computing with Parsytec Computers Gmbh. progressed very well. The group of three people from our Department have visited the production department of the Parsytec Computers, Aachen, Germany for the training and consultations. During this visit the contacts with the Paderborn Parallel Computing Center were established as well.

    3 Work plan

    3.1 Projects

    In 1997 the Department will continue the work on the long-term state programme for the establishing a central federal-level computing and informational resource for the Russian academic and educational network. The Institute will participate in the governmental programme for the creation of high-speed backbone network uniting the most computing and informational centers of the western Russia. The Department will contribute to the SICMA-EAST european programme, that will be done in a close co-operation with St. Petersburg Parsytec division.

    3.2 System integrity and security

    The routine tasks of ensuring system integrity and safety of user's data, making regular backups and performance tuning will be performed. In 1997 we expect the security problem to become one of the most important due to the following factors:

    3.3 User support

    The routine user application, registration, and accounting procedures will be automated. The accounting information being available via hypertext (with access control), as well as the state of the user registration procedure. This will make the registration faster and will allow to eliminate most of the accounting questions. The works on user support will be continued: we plan to organize training courses, deliver lectures, publish new tutorial materials, and provide users with current consultations. The Department will organize the post-bachelor courses for the students of the Institute for Fine Mechanics and Optics on the architecture and programming for parallel systems.

    3.4 Parallel systems

    The Department will continue the investigations of the architecture of the MIMD type. The Institute is going to continue the co-operation with Parsytec Computers Gmbh., that in 1997 will be concentrated on the developing of a software for the Cognitive Computers family.

    3.5 Intranetwork

    The Department is planning to launch an internal project of designing and experimental implementing of the distributed document processing system, integrated with the archive storage system and Institute's library. The system should be (i) network transparent, (ii) platform independent, and (iii) secure.

    3.6 Active network objects

    The initiative on SQL and WWW integration will be extrapolated to cover the active network objects (Java objects).

    3.7 Distributed information retrieval tools

    The opaque caching server will be developed. The integrated indexing and retrieval system is to be designed and implemented, that (comparing to the existing indexing systems) will take the advantage of using all of the information pipelined through the opaque cache server.

    3.8 Networking

    The 1997 year is considered to be a turning point for the networking status of the Institute. We plan to get a direct isolated high-speed connection to the Moscow computer centers. This link will be a prototype of the future backbone network uniting the main computer resources of the european Russia. Thus, the Institute will have to create its own networking control infrastructure.

    3.9 ATM initiative

    The solution of the network problems that the Institute will face in 1997 needs a deeper insight into the modern networking technologies. To be ready to met these demands, we are to organize a special initiative group for the studying of the ATM networking technologies.

    3.10 CADAM group

    In 1997 the Institute plans to organize a class of IBM workstations for the purpose of CAD/CAM development and training. Preliminary studying of the problem is performed by the CADAM initiative group within the Department.
    The file was converted from TeX source with FunnyTeX utility by Mike Krutikov