【正文】 Superdome Sun Fire 25K KBOPS / Core 0 10 20 30 40 Power 595 over million BOPS 50 60 IBM Power 595 3X SPECjbb2023 Java 性能基準測試結果 System (Processor, Cores) BOPS Date. BOPs/JVM Sun Fire E25K ( GHz UltraSPARC VI+, 144core) 2,105,264 May 2023 29,240 HP Superdome ( GHz Itanium2, 128core) 2,054,864 May 2023 64,215 SPECjbb2023** IBM Power 595 ( GHz POWER6 , 64core) 3,435,485 April 2023 107,358 1 產(chǎn)品專家級課程 Page 34 34 Power Systems RAS特性 產(chǎn)品專家級課程 Page 35 AIX 和 POWER6+ 服務器的 RAS 特性 Virtualization ?PowerVM is core firmware ?Dynamic LPAR operations ?No device drivers in PowerVM ?Redundant VIOS support ?Separate HMC Users ?Bare metal Hypervisor ?Live partition mobility General ?First Failure Data Capture ?Hotnode add (570/595) ?Hot node service (570/595) ?Redundant clocks (560/ 570/595) ?Service processors ?Concurrent firmware updates ?CEC bus retry / recovery ?Extensive chip integration reducing ponents CPU ?Dynamic CPU deallocation ?Processor instruction retry ?Alternate processor recovery ?Dynamic processor sparing ?Capacity upgrade ondemand (570/595) ?Processor contained checkstop Memory/Cache ?ECC Chipkill memory ?Storage protection keys ?Dynamic cache deallocation or cache line delete ?Dynamic memory page deallocation ?Spare memory bit steering ?Hardware memory scrubbing I/O ?Redundant I/O links to I/O drawers ?Independent PCI busses ?Dynamic PCI bus slot deallocation ?Hot swap PCI adapters ?Hot I/O drawer add PowerVM Virtual I/O Server LPAR AIX LPAR AIX LPAR AIX LPAR Disk General Memory CPU Network AIX ?Integrated LVM and JFS ?SMIT – reduce human errors ?Dynamic AIX kernel patches ?Concurrent ponent trace ?Live application mobility ?Configurable error logs / boot logs ?Resource monitoring control MTBF in Decades 產(chǎn)品專家級課程 Page 36 Power 595 支持在線維護 CPU節(jié)點 Processor / Memory Node requires service Remove node from running system Repair and Return node System Clock Clock 產(chǎn)品專家級課程 Page 37 Power 595 服務器在線添加 CPU節(jié)點 Add Processor / Memory Nodes having a Without System outage No disruption to running applications No disruption to users. Clock Clock 產(chǎn)品專家級課程 Page 38 Add Processors / Memory Nodes without Rebooting IBM Power 570 企業(yè)級服務器 疊加式設計 ,在線擴展 A. 4 Core to 8 Core B. 8 Core to 12 Core C. 12 Core to 16 Core 產(chǎn)品專家級課程 Page 39 Instruction State Checkpoint GP1=0x14343433 CT=0x12344324 ... RU Core 0 No Error On Fault Soft Error ? Core architected state check pointed every cycle ? Circuitry checked every cycle Non Error Case ? Core restarts from last check point (retry 3 Times reduce 30% CPU failure chance) ? Processor Instruction retry Retries instructions that had faults and protects against intermittent/soft errors Soft Error Case ? Workload is moved to an alternate core ? Alternate Processor Recovery Moves workload over to an alternate/spare processor ? Processor contained checkstops Limits impact of many processor logic/cmd/ctrl errors to just the processor executing the instruction Hard Error Case POWER6 處理器獨有恢復單元 (Recovery Unit / RU) Hard Error Core 1 IFUL S US DUFXURU AltiVecFPUDFUL2Cac h eL2Cac h eL2Cac h eL2Cac h eFPUL S UIFUS DUFXUDFUAltiVecRU產(chǎn)品專家級課程 Page 40 POWER6 透過高能質(zhì)子 (Proton) 射線進行 RAS測試 ? During the Proton Beam experiments 5662 events were recorded. Lead Wall Brass Aperture System Behind Lead Wall % Full Recovery % Resulted in a Partition Outage % Resulted in a System Outage (100 times more failure on Other Processor Chip) 產(chǎn)品專家級課程 Page 41 Feature POWER6 SPARC Integrity Xeon Application/Partition RAS Live Partition Mobility Yes No No Yes Live Application Mobility Yes No No No Partition Availability priority Yes No No No System RAS OS independent First Failure Data Capture Yes No No No Memory Keys Yes No No No Processor RAS Processor Instruction Retry Yes Yes No No Alternate Processor Recovery Yes No No No Dynamic Processor Deallocation Yes Yes Yes No Dynamic Processor Sparing Yes Yes2 Yes2 No Memory RAS Chipkill? Yes Yes Yes Yes Redundant Memory Yes3 Yes3 Yes3 Yes3 I/O RAS Extended Error Handling Yes No No No 服務器 RAS可靠性設計比較 產(chǎn)品專家級課程 Page 42 操作系統(tǒng)停機時間調(diào)查 : AIX是業(yè)界最穩(wěn)定的操作系統(tǒng) (27個國家 400個用戶 ) 012345678910H o u r sWin2023 Win2023 RHEL Solaris HPUX SUSE AIX We are here! 產(chǎn)品專家級課程 Page 43 讓我們共同用 Power 創(chuàng)新企業(yè)未來 ! 謝謝 ! 產(chǎn)品專家級課程 Page 44 ? 靜夜四無鄰，荒居舊業(yè)貧。 , January 29, 2023 ? 雨中黃葉樹，燈下白頭人。 15:19:2315:19:2315:191/29/2023 3:19:23 PM ? 1以我獨沈久，愧君相見頻。 :19:2315:19Jan2329Jan23 ? 1故人江海別，幾度隔山川。 15:19:2315:19:2315:19Sunday, January 29, 2023 ? 1乍見翻疑夢，相悲各問年。 :19:2315:19:23January 29, 2023 ? 1他鄉(xiāng)生白發(fā)，舊國見青山。 2023年 1月 29日星期日 下午 3時 19分 23秒 15:19: ? 1比不了得就不比，得不到的就不要。 。 2023年 1月 下午 3時 19分 :19January 29, 2023 ? 1行動出成果，工作出財富。 2023年 1月 29日星期日 3時 19分 23秒 15:19:2329 January 2023 ? 1做前，能夠環(huán)視四周；做時，你只能或者最好沿著以腳為起點的射線向前。 下午 3時 19分 23秒 下午 3時 19分 15:19: ? 沒有失敗，只有暫時停止成功！。 , January 29, 2023 ? 很多事情努力了未必有結果，但是不努力卻什么改變也沒有。 15:19:2315:19:2315:191/29/2023 3:19:23 PM ? 1成功就是日復一日那一點點小小努力的積累。 :19:2315:19Jan2329Jan23 ? 1世間成事，不求其絕對圓滿，留一份不足，可得無限完美。 15:19:2315:19:2315:19Sunday, January 29, 2023 ? 1不知香積寺，數(shù)里入云峰。 :19:2315:19:23January 29, 2023 ? 1意志堅強的人能把世界放在手中像泥塊一樣任意揉捏。 2023年 1月 29日星期日 下午 3時 19分 23秒 15:19: ? 1楚塞三湘接，荊門九派通。 。 2023年 1月 下午 3時 19分 :19January 29, 2023 ? 1少年十五二十時，步行奪得胡馬騎。 2023年 1月 29日星期日 3時 19分 23秒 15:19:2329 January 2023 ? 1空山新雨后，天氣晚來秋。 下午 3時 19分 23秒 下午 3時 19分 15:19: ? 楊柳散和風，青山澹吾慮。 , Janua