| Cambridge-Cranfield HPCF > Information for Users > Priority Usage |
The Cambridge High Performance Computing Facility (HPCF) provides world class scientific computing facilities to Academic Staff of the University of Cambridge and their Research Teams. For most of the life of the HPCF, the purchase of machines has been funded from competitive funding schemes such as the Joint Research Equipment Initiative (JREI), and more recently the Science Research Investment Fund (SRIF) and the recurrent costs of the Facility have been borne by the users. In the past, users have been encouraged to apply for funding for the HPCF on their research grants but access to the Facility has never been made dependent on payment. However, it is becoming increasingly important that the full costs of running the Facility are recovered. The University has made it clear that it will not cover these costs and it is perfectly conceivable that the electricity supply to the Facility would be disconnected if we do not have the money to pay the bills. The Management Committee have considered a number of mechanisms for recovering the running costs of the Facility placing most emphasis on the central role of the Facility in providing resources for large scale scientific computing. We see no merit in any scheme that would leave the machines unused and so have rejected schemes where all access to the Facility is charged. We are also aware of the small number of staff running the machines and have rejected mechanisms that require large amounts of new software to control queues and access. Instead, we have decided that the best mechanism for encouraging users to contribute to the costs of the HPCF is to provide priority access to the HPCF for researchers whose Project Leaders have contributed to the running costs of the Facility. Subject to the required computing resource being available, any job submitted by a researcher with priority access will run before any other job. Jobs run with priority access will incur a charge for the resources consumed. Once all the funds contributed by a Project Leader have been consumed the relevant researchers will no longer have priority access to the machines.
Details of the initial implementation of this scheme are given below. The priority access scheme will start on 15th October 2002 although Project Leaders will have to take action in advance of this date if they wish their researchers to have priority access from this date. We are aware that there are likely to be difficulties in the practical implementation of the scheme and will modify the scheme as necessary to overcome these. Most importantly, if this charging mechanism fails to raise sufficient income to fund the recurrent costs of the facility it may be replaced by a mechanism where all access to the facility is charged.
The charging regime on the machines in the Cambridge HPCF is aimed at recovering unmet recurrent costs associated with the facility. The costs outlined below are estimates for the average long term recurrent costs under a range of categories. The choice of long term averages is to avoid short term fluctuations in the charging regime as machines come off rolled-up maintenance. These costs are based on the assumption that we will usually have two large machines in the HPCF, one purchased every three years and that each machine will stay in service for six years. These values may be slightly longer than the actual times in practice. However, determining costs over these timescales will allow recurrent costs associated with the Facility to be paid and hence we can guarantee continued provision of computational resources for the length of the longest research grant even if we are unable to attract new funding for replacement machines. While the figures for recurrent costs should not change significantly with time the delivered performance of installed machines is expected to increase significantly with time and this will be reflected in a lower cost per unit of compute following installation of each new machine. This is what has happened previously following the upgrade of the Hitachi SR2201 with the installation of the additional 128 processors in 1997 and again in 2001 following the replacement of the SR2201 with the IBM SP3 system (Hartree). This is also reflected in the significantly reduced cost of Hartree compared to that calculated just a year ago.
Recurrent costs for Cambridge High Performance Computing Facility (Pounds)
Maintenance - (assuming one of the 2 installed machines will be out of the initial rolled-up maintenance period) 100,000 Electricity 200,000 Miscellaneous costs* 40,000 Travel** 10,000 Salaries of 3 Computer Officers (Including 46% overhead) (Grades 1xTop CO I + 2xMid CO II) 160,000 Salary of Administrator (Including 46% overhead) (50% Top CS 5) 16,500 Salary of Accountant (Including 46% overhead) (10% Top CS 5) 3,300 Contribution to refurbishment of computer room and associated services 50,000 Contingency 40,000 TOTAL 619,800 *Miscellaneous costs include: network connection charge, software licences, backup tapes service contracts on air conditioning and alarm system for computer room. ** Travel for computer officers to HPC Conferences, User Meetings and other HPC Facilities.
Expected up time per year (excluding preventative maintenance and
planned downtime for hardware and software upgrades)
8,000 hours per year
The priority access mechanism will prevent optimum scheduling of jobs to
completely fill the machines. It is also important that we avoid
fragmentation of the machines preventing access for large jobs. We
estimate that these effects will mean that 20% of the time on the machines
cannot be utilised.
Utilisable time: 6,400 hours per year
Not all of this utilisable time can be allocated to Research Grants since
a reserve must be retained in order to make good any loss of computer time
due to problems such as power failures, machine failures, data corruption
due to disk failure etc. The reserve will be used as necessary to make up
such loss of charged time.
Time available to charged jobs: 5,760 hours per year
Although Hodgkin (the 64 processor Silicon Graphics) will still be in use on October 15th 2002 this machine could be decommissioned at any time and so it is inappropriate to include it in the charging regime. Thus charges will only be levied for the use of Franklin (the 648-processor Sun and Hartree (the 160-processor IBM SP3). The theoretical peak performance of Franklin is 1890 Gflops and the theoretical peak performance of Hartree is 240 Gflops. Making due allowance for likely node usage patterns and apportioning the total recurrent cost according to single processor performance gives a cost of 14p per processor hour on Franklin and 10p per processor hour on Hartree.
Those primarily responsible for operating the charging mechanism for the HPCF will be the HPCF Administrator (currently Ann Tasker, e-mail admin@hpcf.cam.ac.uk) and the HPCF Computer Officers (currently Nick Maclaren and Paul Smith, e-mail support@hpcf.cam.ac.uk). Copies of all letters and any other paper documents relating to charges should be sent to HPCF Administrator, Centre for Mathematical Sciences, Wilberforce Road, Cambridge, CB3 0WA.
We will not make any attempt to automate the system for gaining access to the high priority queues on the HPCF until a satisfactory mechanism has developed. In the first instance the following mechanism will be adopted. All the paperwork relating to the scheme will be kept by the HPCF Administrator.
In order to initiate priority access to the HPCF Queues, Project Leaders should send an e-mail to admin@hpcf.cam.ac.uk. This e-mail should authorise the transfer of money to the HPCF by specifying the following information
Name of Fundholding Department Title of Project CAPSA Account Research Grant Number and Name (if applicable) Amount to be transferred
Those who have contributed towards the costs of the HPCF prior to 15th October 2002 will be given a credit of value 1/3 of their past contributions to allow priority access for their research teams. A spreadsheet giving information about past contributions to the HPCF can be obtained on request from the HPCF Administrator. Project Leaders should mark this spreadsheet to indicate their past contributions to the HPCF and return it to the HPCF Administrator who will then credit their account with 1/3 of their past contributions.
Project Leaders should e-mail the HPCF Administrator to indicate which researchers are to be allowed priority access to the HPCF in return for the payment made. Project leaders can choose to allow a number of researchers to spend against the sum transferred OR to apportion the funds transferred to allow a spending limit per researcher.
Once the money has been paid into the HPCF Funds, the HPCF Administrator will ask the HPCF Computer Officers to activate priority access for the researchers specified by the Project Leader.
If Project Leaders wish to remove a researcher from the list of those authorised for priority access they should send an e-mail to this effect to support@hpcf.cam.ac.uk (with a copy the admin@hpcf.cam.ac.uk). It is expected that such requests will be implemented within one working day but this cannot be guaranteed.
It is the responsibility of Project Leaders to brief members of their Research Teams on the charging mechanism and on the consequences of having priority access to the HPCF. In particular they must ensure that members of their Project Teams are aware that once authorised for priority access ALL use of the HPCF will incur charges.
Every month the HPCF Computer Officers will notify the HPCF Administrator of the machine usage by researchers with priority access to the HPCF. The HPCF Administrator will update a spreadsheet of the funds allocated to the HPCF against charges incurred and will send a copy of the relevant entries to each Project Leader who has transferred money to the HPCF.
Notification of any loss of computer time or incorrect charging should be made direct to Mike Payne - mcp1@phy.cam.ac.uk (with a copy to the HPCF Administrator).
The HPCF Management Committee is aware of a number of potential
difficulties with the charging scheme. For instance some researchers may
be involved in several projects and as a result might have more then one
Project Leader. Any Project Leader who feels that they or their Research
Teams will be unduly restricted or greatly inconvenience by the suggested
charging mechanism is invited to discuss their concerns with Mike Payne.
M.C. Payne
23rd September 2002.