NIKHEF Test Bed Status David Groep davidg@nikhef. nl
NIKHEF: Current Farms and Network STARTAP 2 x 622 Mbit/s 2. 5 Gb/s SURFnet NREN (10 Gbit/s) NIKHEF Edge Router STARLight & CERN both 2. 5 Gb/s IPv 6 1 Gb IPv 4 1 Gb Farm. Net “backbone” – Foundry 15 k Development Test Bed 5 x dual-PIII FNAL/D 0 MCC 50 x dual-PIII Application* Test Bed 20 x dual-PIII NCF GFRC+ DAS-2 Cycle Scavenging 32 x dual-PIII 168 x dual-PIII 60 x dual-AMD David Groep – NIKHEF Test Beds – 2002. 08. 26 - 2
Test Bed Buildup stategy “Why buy farms if you can get the cycles for free? ” u u u Get lots of cycles in “scavenging” mode from CS research clusters Attracts support from CS faculties Get cycles from national supercomputer funding agencies Downside: u Many different clusters (but all run Globus and most EDG middleware) u Middleware shall (and should) be truly multi-disciplinary! David Groep – NIKHEF Test Beds – 2002. 08. 26 - 3
SARA: Mass Storage u NIKHEF “proper” does not do mass storage – only ~ 2 TByte cache u SARA: 200 Tbyte Storage. Tek Near. Line robot u 2 Gbit/s interconnect to NIKHEF u Front-end: “teras. sara. nl” 1024 processor MPP – SGI IRIX u Ron Trompert ported GDMP to IRIX. Now running! David Groep – NIKHEF Test Beds – 2002. 08. 26 - 4
Challenges and Hints u Farm installation using LCFG works fine n n u Re-install takes 15 minutes (largely due to application software) Adapts well to many nodes with different functions (2 x. CE, 2 x. SE, 2 x. UI, external disk server, 2 acceptance-test nodes, 2 types WN, D 0 nodes, …) Some remaining challenges n “edg-release” configuration files are hard to modify/optimize n Red. Hat 6. 2 is really getting old! n Netbooting for system without FDD n Get all the application to work! David Groep – NIKHEF Test Beds – 2002. 08. 26 - 5
LCFG configuration u Use EDG farm to also accommodate local user jobs u disentangled hardware, system, authorization and app. Config u modified rdxprof to support multiple domains u using autofs to increase configurability (/home, GDMP areas) u Installed many more RPMs (DØMCC, LHCb Gaudi) and home-grown LCFG objects (pbsexechost, autofs, hdparm, dirperm) u Force u Shows RPM install trick (+updaterpms. offline) flexibility of LCFG (with PAN it will be even nicer!) David Groep – NIKHEF Test Beds – 2002. 08. 26 - 6
Red. Hat 6. 2 – modern-processor breakdown u Recently acquired systems come with P 4 -XEON or AMD K 7 “Athlon” u Kernel on install disk (2. 2. 13) and in RH Updates (2. 2. 19) say “? ? ? ” u Baseline: Red. Hat 6. 2 is getting really old u But a temporary solution can still be found (up to kernel 2. 4. 9): use new kernel (without dependencies) in existing system u Requires you to build a new RPM u You can even get the Intel 1 Gig card to work (after install only *) u See http: //www. dutchgrid. nl/Admin/Nikhef/edg-testbed/ David Groep – NIKHEF Test Beds – 2002. 08. 26 - 8
Installing systems without an FDD u Most modern motherboards support PXE booting u stock LCFG-install kernel works well with PXE u “just” need a way to prevent an install loop n thttpd daemon with a perl script to “reset” dhcpd n called from modified dcsrc file n script will only reset dhcpd. conf when $REMOTE_ADDR matches n CNAF did something similar using temporary ssh keys David Groep – NIKHEF Test Beds – 2002. 08. 26 - 9
Our test bed in the Future u We expect continuous growth u Our Aims: u u “infinite” storage @ SARA u 2. 5 Gbit/s interconnects now u u ~ 1600 CPUs by 2007 > 10 Gbit/s in 2003/2004? Our constraints: u The fabric must stay generic and multi-disciplinary David Groep – NIKHEF Test Beds – 2002. 08. 26 - 10
Скачать презентацию NIKHEF Test Bed Status David Groep davidg nikhef nl
5497d3e86b0f9e6873b31db2f3fd8276.ppt