TSUBAME service in FY2018 has opened

April 4, 2018

TSUBAME service in FY2018 has opened. The changes are as follows:

  1. Point calculation formula has changed
  2. Storage Service Terms of Usage added
  3. Restriction on job execution added
  4. The default version of the applications has changed
  5. Project quota started
  6. SSD file format has been changed
  7. Reservation service resumed
  8. Improved connection performance to compute nodes and external networks
  9. Acceptance of the application for use by research based on competitive funds and collaborative research agreement started
  10. Resource type s_gpu is now available

1. Point calculation formula has changed. Please refer to 規則・利用細則 for details.

2. About the storage service provided in TSUBAME3.0, "東京工業大学学術国際情報センターストレージサービス利用規約" was newly enacted. Please see 規則・利用細則 for details.

3. The following restrictions have been applied to prohibit specific users from occupying the computing resources.

 Number of concurrently executable jobs per user: 150
 Number of concurrently executable slots per user: 2016

4. The default version of the applications has changed as follows.


Last year         This year
cuda/8.0.44       cuda/8.0.61
intel/  intel/
pgi/17.5          pgi/17.10


Last year           This year
intel-mpi/17.3.196  intel-mpi/18.1.163
openmpi/2.1.1       openmpi/2.1.2


Last year                 This year
forge/7.0.5               forge/18.0.1
intel-ins/   intel-ins/       
intel-itac/17.3.030       intel-itac/18.1.017
intel-vtune/ intel-vtune/


Last year                 This year
ansys/R18.1               ansys/R18.2
comsol/53                 comsol/53a_u2
gaussian16/A03            gaussian16/B01
gaussian16_linda/A03      gaussian16_linda/B01
mathematica/11.1.1        mathematica/11.2.0
matlab/R2017a             matlab/R2017b

5. The quota of the group disks has changed to project quota. Previously, the total size of a group disk was measured by only the files of which owning group is the group, but currently the total size of it is calculated regardless of the owning group of the file. This change makes it easier to see which group's capacity will be charged when moving files between groups.

6. To avoid bugs of xfs file system, SSD (NVMe) of each compute node was fotmated from xfs to ext4.

7. The node reservation function had been suspended temporarily because the reserved node could not be used normally due to a problem of the scheduler, but the problem was corrected and it is now resumed. Please not that there is a possibility that the reservation function will be suspended without prior notice in the event that the scheduler problem recurs.

8. When accessing the Internet from a compute node, there was a problem that network bandwidth could not be fully utilized. This bug has been corrected and it has become possible to connect with the Internet environment including inside and outside the campus with broadband.

9. See here in detail.

10. The bug of the s_gpu was fixed by version-up of the batch job scheduler and it becomes available.