Introduction

Several considerable facts might be taken into account:

• decreasing the financial resources in HEP;
• increasing the requirements for computing;
• existing of a cheap operating platform like Linux;
• cheap commodity hardware (PC);
• existing batch systems CODINE, NQS, LSF, and others which are relatively easy available in many leading HEP laboratories;
• almost all important applications in HEP might be running on Linux;
• new languages: Java;

We might consider General Purpose Computing Facility on a base of PCs.

It is discussed PC farm consisting of N nodes (PCs) for batch processing with relatively weak communications between nodes. N might be considered in range from several PCs to probably 100 or more. There is already known computing system based on 9K (!) processors PentiumPro200 (Intel).

Reliability of an existing hardware is enough to support PC farm consisting of 100 nodes by one person (for hardware, operating system and general application software).

Scalability: we can increase a number of PCs to desired value in step by step mode by small quanta.

The value about 400 Pentium Pro 200 MHz or 200 Pentium II 300 MHz can be discussed as really achievable maximum of computing power (in case when four processor motherboards will be used).

One processor Pentium II 300 MHz is more than two times more powerful the R4400 200 MHz. That means the maximum computing power will be more than 30 SGI with 28 R4400 processors every SGI. In other words Pentium II 300 MHz has about 11 SpecInt95 and about 8 SpecFp95 (about 100 MFlops in Linpack).

A number of SCSI ports and disk drives may be various and will reflect concrete needs in the discussed proposal.