2. A UK Computing Facility John Gordon RAL

3. October ‘99HEPiX Fall ‘99 Data Size Event Rate 10 9 events/year Storage Requirements (real & simulated data) ~300TByte/year UK Physicists want access to data for analysis 2TB in 1999, 4TB/year 2000 on data and simulation

4. October ‘99HEPiX Fall ‘99 Why don’t UK Physicists use SLAC? ….and SLAC is already heavily used.

5. October ‘99HEPiX Fall ‘99 Existing UK Facilities Shared facilities in UK are HP, Intel/Linux and Intel/NT. BaBar mainly use Suns –Historically, UK lacking in Suns in HEP departments BaBar have Sun E3500, 4 cpus, 2GB of memory at RAL - Bought for program development Several hundred GB of disk Plus a few desktop machines in universities

6. October ‘99HEPiX Fall ‘99 BaBar bid for more BaBar went to a UK Government research fund and bid for $1.8M for UK BaBar facilities They were awarded ~$1.2M at the start of this year for: –A central server at RAL with several TB which will receive data from SLAC. –Server and disk in 10 UK universities –Co-operating databases across the UK –One extra staff member to achieve this

7. October ‘99HEPiX Fall ‘99 Actual Equipment Sun vs Compaq Sun won. RAL 5 Universities (Bristol, Edinburgh,Imperial, Liverpool, Manchester) with 4 Universities (Birmingham, Brunel, QMW, RHBNC) with E4500 server - 6x400MHz cpus, 4GB memory 5TB of formatted disk in 27 A1000 RAID arrays 6 UWSCSI busses DLT7000 stacker 7 fast ethernet adaptors E250, 2 x400MHz cpus, 1GB, 3xA1000 = 0.5TB E450, 3x400MHzcpus, 2GB 5x A1000, = 1TB

8. October ‘99HEPiX Fall ‘99 Setup at RAL (early experience) Equipment delivered and installed Filesystems limited to 1TB –used 4xA1000 => 720GB striped(?) 5.5 Million events brought from SLAC E3500 acts as a front-end, E4500 holds data, both runs batch jobs E4500 also AMS server to other systems. LSF cluster on 2 Suns. Who else is running large data on Suns?

9. October ‘99HEPiX Fall ‘99 OOSS Andy Hanushevsky visited in September and installed his OOFS and OOSS This provides a layer which interfaces Objectivity to the Atlas Datastore (cf HPSS at SLAC) All the disk space runs under the control of OOS which acts as a cache manager Current level of Objectivity/AMS doesn’t allow OOS to retrieve data transparently from the robot but data can be easily brought on-line by prestaging

10. October ‘99HEPiX Fall ‘99 Network Plans A single server in a university on fast ethernet can suck data from RAL at rates which will be unpopular with other sharing the institutes connections to the WAN. Pilot to establish tunnels over JANET using spare ATM capacity

11. October ‘99HEPiX Fall ‘99 Manchester RAL IC 2MB

12. October ‘99HEPiX Fall ‘99 Purpose of Trial Since bandwidth will be small the trial will not necessarily give better throughput Establish whether end-to-end connection over PVCs works Establish whether the different management domains can reach a common, working solution Check that the routing works Should be simple to increase bandwidth later

13. October ‘99HEPiX Fall ‘99 Original Data Model Data to RAL by tape Model I - all TAG data at other sites; pull detailed data from RAL Model II - frequently-accessed events stored in full at other sites; replication from RAL Investigate methods of copying, updating, replicating databases over WAN

14. October ‘99HEPiX Fall ‘99 New Data Model(?) BaBar currently has performance limitations Working on non-Objectivity solution NOTMA DST using ROOT i/o 10**9 events = 280GB Likely that universities will want all events locally Detailed events stored at RAL in Objectivity

15. October ‘99HEPiX Fall ‘99 Conclusion RAL moving in opposite direction from HEPCCC proposal - more flavours of unix on new hardware platform. BaBar will be using Linux soon for simulation though A bigger scale of disk data handling for one experiment. Data synchronisation over WAN –(SLAC-RAL-UK Universities)