Upgrade to Pro — share decks privately, control downloads, hide ads and more …

SSII2026 [OS1-2] 学術クラウド基盤mdx IIの 設計と運用

SSII2026 [OS1-2] 学術クラウド基盤mdx IIの 設計と運用

More Decks by 画像センシングシンポジウム

Transcript

  1. ࣗݾ঺հ w ໊લߴڮܛஐ ͔ͨ͸͚͍ͪ͠  w ॴଐେࡕେֶ%ηϯλʔ εύίϯ౳ͷશࠃͷݚڀऀ͕ڞಉར༻͢Δ ܭࢉࢿݯͷઃܭɼௐୡɼӡ༻ 

    w ઐ໳ߴੑೳܭࢉ ಛʹεύίϯͷγεςϜ ιϑτɼϓϩάϥϛϯά؀ڥɼੑೳ࠷దԽͳͲ  w ܦྺ w େࡕେֶ ʙ೥  w ಸྑઌ୺Պֶٕज़େֶӃେֶॿڭ ʙ೥  w ౦๺େֶαΠόʔαΠΤϯεηϯλʔॿڭ ʙ೥  w େࡕେֶ%ηϯλʔ।ڭत ೥݄ʙ 2
  2. NEY w େֶɾݚڀػ͕ؔڞಉௐୡɾӡ༻͢ΔΫϥ΢υج൫ ⿞ॊೈͳܭࢉ؀ڥͷߏங ⿞҆શͰִ཭͞ΕͨԾ૝؀ڥͷߏங ⿞σʔλιʔεͱͷγʔϜϨεͳ࿈ܞ ⿞ϦΞϧλΠϜۓٸδϣϒ΁ͷରԠ w NEY*͸೥ʹ౦ژେֶʹಋೖ 4

    NEY*ͷ֓ཁ<> T. Suzumura et al., “mdx: A Cloud Platform for Supporting Data Science and Cross-Disciplinary Research Collaborations,” DASC/PiCom/CBDCom/CyberSciTech, 2022.
  3. NEY**ͷΞʔΩςΫνϟ 8 CPUϊʔυ x60 CPU CPU NIC RAM RAM GPUϊʔυ

    x15 CPU CPU NIC RAM RAM GPU GPU NIC Ethernet 200Gbps GPU GPU NFS S3DS Nextcloud Object storage 432TB Lustre 553TB ؅ཧαʔό *OUFSOFU 4*/&5
  4. ܭࢉϊʔυͷߏ੒ 10 $16ϊʔυ (16ϊʔυ CPU Intel Xeon Platinum 8480+ (56

    cores) x2 Intel Xeon Gold 6530 (32 cores) x2 Memory 512 GiB (DDR5-4800 SDRAM ) 1024 GiB (DDR5-5600 SDRAM ) GPU N/A NVIDIA H200 SXM5 x4 Network 200 Gbps Ethernet x1 200 Gbps Ethernet x2 # of nodes 60 15 $16ϊʔυ ϊʔυ6 (16ϊʔυ 6
  5. ετϨʔδઃܭ 11 ΠϯλʔϑΣʔε ༰ྔ ໨త Block POSIX 100 TB 7.σΟεΫ

    Lustre POSIX 1,006 TB ߴੑೳɾฒྻ*0 S3DS S3-compatible API ߴੑೳ*0 HyperStore S3-compatible API 432 TB σʔλऩ༰ɾΞʔΧΠϒɾެ։ /dev/vda /lustre /nfs WJSUJPCML )PTU (VFTU -VTUSF /'44FSWFS 4%4 4FSWFS &9"4DBMFS )ZQFS4UPS F 4 4 &9"4DBMFS্ )ZQFS4UPSF্
  6. -VTUSFͷར఺ɾܽ఺ w εύίϯͳͲେن໛ΫϥελͰ޿͘༻͍Β Ε͍ͯΔฒྻ෼ࢄϑΝΠϧγεςϜ w ༰ྔ΍ੑೳ͸΄΅ແ੍ݶɺ଱ো֐ੑ΋ߴ͍ w 104*9ޓ׵ͷͨΊɺΞϓϦͷमਖ਼͸ෆཁ w ઃܭ΍ӡ༻͕೉͘͠ɺ044Ͱ͸͋Δ͕ϕϯ

    μ %ࣾ ͷӡ༻อक͕࣮࣭తʹෆՄܽ w )1$؀ڥ͔Βൃల͖ͯͨͨ͠ΊɺϚϧνς φϯγͳͲΫϥ΢υ޲͚ػೳ͸΍΍ऑ͍ 14 IUUQTXJLJMVTUSFPSHΑΓ