3rd International WWW Caching Workshop, Manchester, June 15-17 1998


WWW Caching in Russia - Current State and Future Development

Serge KRASHAKOV and  Lev SHCHUR

Landau Institute for Theoretical Physics
142432 Chernogolovka, Russia
tel/fax: +7 (095) 913-2317
e-mail: sakr@chg.ru

Our main goal is the creation and development  of WWW caching for Research and Education sector of Russian networking under financial support via grants 96-07-89266 and 98-07-90293 from Russian Foundation for Basic Research (RFBR). So, here we pay attention mainly to the sector of R&E networking in Russia.

1. Introduction.

History of R&E networks in Russia

History of a wide-area IP networking in Russia started in 1990-1991. RELCOM as public (and commercial service provider) and FREEnet - Network For Research, Education, and Engineering - was the first ones. Nowadays, FREEnet is the most developed network for science and education in Russia connecting more than 350 scientific and educational institutions in 17 regions, such as Moscow, Chernogolovka, Yaroslavl, Chelyabinsk, Perm, and many others, located from Central Russia to Siberia. The others R&E networks are RBnet, RUNnet, RSSI-Net and Radio-MSU.

Networks Interconnection

Practically all Russian networks are interconnected via Internet eXchange points in Moscow (M9-IX), St.Petersburg and via South Moscow Backbone at 10 Mbps. The following picture roughly shows the major Russian networks and their connectivity. Practically all major russian networks has own international channels with bandwidth from 256 kbps to 6 Mbps.


Regional Interconnection (Inner network links)

Regional networks in the past (and some of them even now) was connected with Moscow by analogue 4-wire leased lines with bandwidth 19.2-28.8 kbps. Regional network in Chernogolovka - Chg-FREEnet was the first who began to take some steps to reduce their external traffic. First solution was to set up a ftp and www mirroring of most interesting sites.

First steps with caching

We start our caching server in Chernogolovka in October 1995 after appearance of Harvest Proxy Cache. The first steps in creating national caching hierarchy began in September 1996 when central FREEnet cache server was installed in Zelinsky Institute. Practically all regional FREEnet networks cache-servers began to use it as their parent. At the same time or later neighbor-neighbor relationships with cache servers from some other Russian networks (RSSI-Net, ITEP-Net) was established. Since February 1997 we started to establish sibling relations with our nearest foreign neighbors - salvator.ecrc.de (Munich, ECRC) and ebone-proxy.univie.ac.at (Vienna, EBONE).

2. Current State.

Number of cache-servers in Russian organizations grows quite rapidly. Many networks and ISP set up central cache-servers. And until now, all cache-servers was built on internal means of organizations without any special funding from federal institutions and this was the main problem of relatively slow development of large-scale infrastructure.

In our knowledge, FREEnet has the most advanced mesh of cache-servers. Cache servers are installed in all regional FREEnet branches and have peering with central FREEnet cache server. After construction of 2 MBps line Chernogolovka-Moscow, www-cache.chg.ru located in Chernogolovka is the main FREEnet cache-server de-facto.

Central cache-servers of many networks have neighbor-neighbor internetwork relations. In a framework of TERENA TF-CACHE international activity on creating European cache-mesh we have peering agreement with some foreign neighbors: FREEnet cache-servers webcache1.free.net and www-cache.chg.ru - with salvator.ecrc.net (Munich,ECRC), ebone-proxy.univie.ac.at (Vienna, EBONE), and sd.cache.nlanr.net (San Diego, NLANR). Radio-MSUnet cache-server alpha1.radio-msu.net, and ITEP server cache.itep.ru - with DFN cache in Hamburg etc. The next figure shows current mesh of cache-servers in Russia:


3. Some facts

Typical regional server has about 100-200 k requests/per day from about 500 clients, transfers about 500 MB - 1 GB per day, and %hit is about 40-60%. Typical central cache-server of a network has about 500-1000 k requests per day (total sum of ICP-requests from neighbors and HTTP-requests from neighbors and HTTP-clients), transfers from 2 to 6 GB per day, %hit is about 30% for ICP-requests and about 40-50% for TCP-requests.

At the moment, the most popular hardware for building cache-servers is the fast PC (AMD K6, Pentium Pro, or Pentium-II with 64-192 MB RAM, and 4-12 GB Fast-SCSI disks) operating under FreeBSD or Linux. Some places prefer SUN UltraSparc with 128-256 MB RAM.

The most popular caching software - Squid.

Fresh Winter Olympic example

Working with squid cache servers discovered advantages of their using with slow and unreliable communication lines: very often it is impossible to receive information directly because of big amount of lost packets, but at the same time information may be received via the chain of cache servers (hop by hop). The latest and striking example is the Winter Olympic games in Nagano when very often direct connection of FREEnet with Nagano was broken and we use to receive information our parent cache in San Diego.

4. Perspectives

Recently, our proposal to Russian Foundation of Basic Research on the creating national caching system for science and education was approved as grant number 98-07-90293. Proposal assumes installation of several powerful cache servers in some of the key points of network (M9, FREEnet NOC, Institute of Space Research, Prezidium of RAS, etc). Procedure of coordination between web-cache administrators of all R&E networks is also proposed. We expect to introduce step by step the source cache routing to equilibrate international and interregional traffic of scientific networks.

Clearly, extensive contacts and cooperation with foreign national cache programs are of great importance for us.

5. Problems

Having the cache server means that you solved some problems, but the new ones comes out as well. Generally this problems may be divided into 2 categories: technical and organizational, and very often organizational problems are much more difficult to solve rather than technical ones.

Proxy server is a bottleneck of network access. If it is down the whole network is down. High quality of service needs high level of redundancy and back-up facilities. Therefore, "corresponding" level of financing is needed.

Among unsolved problems are very unstable routing. We think to use source cache routing to determine the nearest neighbor with the best connectivity to source of informations. Fortunately, recently, we found growth of interest from financial organizations to the R&E Internet development. This gives to us hope that overall International bandwidth of networks in Russia will be doubled to the end of this year.