Golias farm

The Golias farm uses the HTCondor batch system for distributing jobs to individual computing servers. HTCondor servers accept jobs from local and grid users and distribute them to worker nodes. Worker nodes are divided by hardware type into subclusters: Milan, Minis, Mahagon and Mikan.

Overview table of subclusters and their main parameters (July 2023):

subcluster# servers# cores# HT coresRAM [GB]power [kW]performance [HS06]HS06/W
Milan (2022)291 8563 71214 84821.766 0733.0
Minis (2020)121923847682.44 0241.7
Mahagon (2019)247681 5363 0728.718 0632.0
Mikan (2017)511 2242 4486 52817.526 8541.5
in total1164 0408 08025 21650.3115 0148.2

Milan subcluster

8x Gigabyte H262-Z6B-00 2U size chassis, each includes 4 servers; in total 29 computing nodes. Each node is equipped with two 32-core processors of type AMD EPYC 7543, 2.8GHz, 512 GB DDR4 RAM and 3.84 TB Samsung SSD. Performance per server 2278.4 HS06, HT enabled. The measured power consumption of the entire subcluster at maximum load is 21.7 kW. In operation from April 2022.

Total subcluster performance: 66073 HS06 (3712 cores with HT enabled).

Minis subcluster

12x ASUS RS500A 1U size server. Each server is equipped with one 16-core processor type AMD Epyc 7351P, 2.4GHz , 32 GB DDR4 RAM and 480 GB SSD Kingston DC500M. The measured power consumption of one server is 200 W. In operation since the end of December 2020.

Total subcluster performance: 4024 HS06 (384 cores with HT enabled).

Mahagon subcluster

6x HPE Apollo 35 2U size chassis, each includes 4 servers; in total 24 computing nodes. Each node is equipped with two 16-core processors of type AMD EPYC 7301, 2.2GHz , 128 GB RAM DDR4 and 1.92 TB SSD Intel S4510. Performance per server 752 HS06, HT enabled. The measured power consumption of the entire subcluster at maximum load is 8.7 kW. In operation since May 2019.

Total subcluster performance: 18063 HS06 (1536 cores with HT enabled).

Mikan subcluster

51x Huawei RH1288 V3 1U size server. Each node is equipped with two 16-core processors of type Intel Xeon E5-2650 v4, 2.2GHz , 128 GB DDR4 RAM and two 600 GB SAS drives. Power 526 HS06 per server, HT enabled. Measured power consumption of one server at maximum load is 342 W, the entire subcluster 17.5 kW. The subcluster was acquired from the funds of OP VVV projects: CERNC, AUGER.CZ and Fermilab – CZ. In operation since November 2017.

Total subcluster performance: 26854 HS06 (2448 cores with HT enabled).

MFF subcluster

It consists of various worker nodes, located at Troja Campus. Nodes are connected directly to the farm router via a dedicated line CzechLight.

  • mff1701-mff1704 – servers purchased in December 2017 – Supermicro X11SSL-F, CPU Intel Xeon E3-1270 v6 @ 3.80GHz (8-core), RAM 16GB DDR4 2400MHz, HDD 0.5TB, HEPSPEC06 16.96/core (∑ 543 HEPSPEC06)
  • mff1801-mff1804 – servers purchased in December 2018 – Supermicro X11SRM-F, CPU Intel Xeon W-2125 CPU @ 4.00GHz (8-core), RAM 32GB DDR4 2666MHz, HDD 1TB, HEPSPEC06 18.75/core (∑ 600 HEPSPEC06
  • mff1901-mff1903 – servers purchased in December 2019 – Supermicro X11SRM-F, CPU Intel Xeon W-2125 CPU @ 4.00GHz (8-core), RAM 32GB DDR4 2666MHz, HDD 1TB, HEPSPEC06 18.75/core (∑ 450 HEPSPEC06)
  • mff2001-mff2007 – servers purchased in 2020 – Supermicro X11DDW-L, 2x CPU Intel Xeon Silver 4216 CPU @ 2.10GHz (2x 32-core), RAM 196GB DDR4 2933MHz, SSD 2TB, HEPSPEC06 11.30/core (∑ 5062 HEPSPEC06)
  • mff2101-mff2104 – servers purchased in December 2021 – Asus RS700A-E9-RS4V2, 2x CPU AMD EPYC 7352 @ 2.3GHz (2x 24-core, HT on), RAM 256GB DDR4 3200MHz, SSD 2TB, HEPSPEC06 13.47/core (∑ 5172 HEPSPEC06)
  • mff2201-mff2203 – servers purchased in March 2022 – Asus RS700A-E9-RS4V2, 2x CPU AMD EPYC 7352 @ 2.3GHz (2x 24-core, HT on), RAM 256GB DDR4 3200MHz, SSD 2TB, HEPSPEC06 13.47/core (∑ 3879 HEPSPEC06)
  • mff2301-mff2303 – servers purchased in 2023 – Gigabyte H242-Z11-00, 1x CPU AMD EPYC 7352 @ 2.4GHz (1x 32-core, HT on), RAM 128GB DDR4 3200MHz, SSD 2TB, HEPSPEC06 13.9/core (∑ 2669 HEPSPEC06)

Golias Farm Overviews:

en_USEnglish
Přejít nahoru