UW CMS Tier-2 Resources

Computing

The High Energy Physics group at the University of Wisconsin strives to maintain an information technology infrastructure that is trouble-free, secure, highly available and well understood. The table below provides a summary of the expected available computing (and storage) resources. In practice, due to disk failures and lack of back up disks the total storage can be less than or equal to the number shown.

GenCoresCPU Class (year puchased)Slots*HS06/slot**Total HS06Storage (TB)GPU
g26322.2GHz Xeon E5-2660 (2013 Winter)3849.643702480
g27402.2GHz Xeon E5-2660V2 (2013 Fall)12009.6411568320
g28402.3GHz Xeon E5-2660V2 (2014 Fall)13609.6413110346
g29402.3GHz Xeon E5-2650V3 (2015 Fall)116010.8812615316
g30402.3GHz Xeon E5-2650V3 (2016 Mar)120010.8813050477
g31482.2GHz Xeon E5-2650V4 (2016 Fall )326410.21333201067
g32482.2GHz Xeon E5-2650V4 (2017 Fall )96010.2198001194
g33482.2GHz Xeon E5-2650V4 (2018 Fall )81610.2183301738
g34482.8GHz AMD EPYC 7402P (2019 Fall )67214.0694501818
g35642.3GHz Xeon Gold 5218 (2019 Fall )6411.597420
g36643.0GHz AMD EPYC 7302 (2020 Fall )134415.5208322600
Total–1242410.99 (Wtd)136519TBS

*: Slot counts are working batch slots as of Mar 30, 2021.

**: HS06 benchmark was run with all services disabled other than sshd, afsd, and xinetd. The following software was used:

GCC: gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-52)
C++: g++ (GCC) 4.1.2 20080704 (Red Hat 4.1.2-52)
FC:  GNU Fortran (GCC) 4.1.2 20080704 (Red Hat 4.1.2-52)
SPEC2006 version: 1.2

We use Condor batch computing software to implement a high throughput Linux computing environment. Opportunistic computing resources from the Wisconsin CMS T2 (aka Grid Laboratory of Wisconsin (GLOW)) and UW’s Center for Highthroughput Computing (CHTC) provide the potential for the utilization of a total of over 17,500 Linux CPUs.

Network

Compute nodes from prior to 2017 have 1G links to our LAN. All newer ones have 10G. The 10G endpoints are connected to Nexus 93108TC-EX rack switches. These rack switches each have 100G links to a pair of Nexus 9336C-FX2 building switches that together are connected at 2x100G to the campus research network backbone, which operates on 4x100G links. The UW–Madison research backbone is connected to wide area research networks in Chicago via two independent routes each of which is composed of 2x100G links. In Chicago, the connections are to national research networks including Internet2 and ESNet. Via ESNet, our CMS Tier-2 is connected to the LHCONE international network.

[Research Backbone Network Diagram]

Storage

The storage system is based on HDFS with an xrootd interface. A total storage space of more than 4PB is distributed across many dedicated and dual-use commodity servers in the cluster. To avoid data loss when hard disks fail, two copies of each HDFS block is maintained, so the amount of data that can be stored is half the total disk space.