Azul Java

Transcription

1 Azul Systems - Java 유료화대응방안 - 권범준, Azul Systems 한국대표 joon@azul.com Copyright Azul Systems Azul Systems 2016

2 2 Copyright Azul Systems Azul Systems

3 3 Copyright Azul Systems Azul Systems

4 4 Copyright Azul Systems Azul Systems

5 Java SE Version Java SE Lifecycle Historical 1-year Public updates 4-month Public updates overlap overlap 11 (LTS) 11 (LTS) 1.7-year Public Oracle acquires updates Sun (Jan, 2010) 8 (LTS) overlap 8 (LTS) 3-year Public updates 7 (LTS) overlap 7 (LTS) 6 (LTS) 6 (LTS) 5 (LTS) 5 (LTS) Publicly available binaries (unsupported) Oracle Commercial Support Oracle Extended Commercial Support 5

6 Java SE Version Java SE Lifecycle 5+ Year Timeline (LTS) Public updates zero overlap 17 (LTS) 16 ( Support Cliff ) (LTS) 11 (LTS) Note: Oracle information from (LTS) 8 (LTS) 6 7 (LTS) 7 (LTS) Oracle Publicly available binaries (unsupported) Oracle Commercial Support Oracle Extended Commercial Support

7 Java SE Version Java SE Lifecycle: 5+ Years (LTS) 16 Azul Zulu LTS Support 8 years active (LTS) Azul Zulu MTS Support Bridge to next LTS (LTS) 11 (LTS) (LTS) 8 (LTS) 7 (LTS) Oracle Publicly available binaries (unsupported) 2019 Oracle Commercial Support Oracle Extended Commercial Support 2022 Azul Zulu Community Builds Azul Zulu Production Commercial Support 7 (LTS) Azul Zulu Extended (Passive) Commercial Support 7

8 Easy Migration from Oracle JDK 8 Copyright Azul Systems Azul Systems 2016

9 Why Migrate Your Java?

10 10 기존 JDK 버전공개정책 충분한기간을제공하던기존버전공개정책 JDK 6: 2 년 11 개월 (from public JDK 5 updates) JDK 7: 1 년 9 개월 (from public JDK 6 updates) JDK 8: 1 년 1개월 (from public JDK 7 updates) 새플랫폼 ( 버전 ) 의안정적인상용화적용을위한오버랩기간제공 10

11 11 Open JDK: 새로운버전공개정책 새로운 JDK 버전공개정책 : 매 6 개월 (A new version of the JDK is being released every six months ) - 매년 3 월, 9 월새버전출시 11

12 새로운 JDK Binary 배포및업데이트정책변화 새 OpenJDK 는다음버전출시되는 6 개월동안만업데이트제공 (will only be updated for six months until the next JDK is released) 그중오라클의 LTS 가제공되는버전만계약하에기술지원가능 (Long-term support(lts) from Oracle will only be available for the Oracle JDK binary under the commercial support agreement) JDK 8 공개업데이트는 2019 년 1 월까지제공됨 (Public updates for JDK 8 end in January 2019) 년 1 월이후, 기업사용자는현재처럼 JDK 8 을사용할수있으나, 보안패치나오류수정업데이트는받을수없음 (After that, users will still be able to use JDK 8 in the same way they have previously, but there will be no more security patches or bug fixes)

13 변경되는 Binaries 내용 Oracle JDK OpenJDK Java SE OpenJDK Oracle JDK Azul JDK Java SE JDK 11 and later JDK 10 and earlier 13

14 14 변경되는 Binaries 내용 (JDK 11) 기존제공되던일부 Binaries 는오픈소스화됨 (will be open-sourced) Flight recorder (JDK 11) Mission control Others 기존제공되던일부 Binaries는삭제됨 (will be removed) Browser Plugin Java Web Start JavaFX 14

15 15 정확한정책해석 : Oracle JDK vs. Open JDK Oracle JDK (from JDK 11) 는상용계약고객에한하여사용가능 (ONLY be used in production with a commercial support contract) 계속무상사용하고자하는기업고객은 OpenJDK 만사용가능 (The only free JDK 11 and later will be OpenJDK binaries) 그러나, OpenJDK 를사용하면서계속무상으로보안패치및오류수정업데이트혜택을받고자하는기업고객은매 6 개월마다새로운 JDK 버전으로업그레이드해야함. (To continue to receive free updates to the JDK you MUST update your JDK EVERY SIX MONTHS) 15

16 Example Customer case In Java Decision

17 기업고객사례 (Example Customer case) 2018 년현재, 전세계에서 60,000 이상의서버를운영중인기업 오라클의새정책에따라 2018 년말까지의사결정목표설정 향후 Java 정책에대하여 4 가지옵션최종평가 (1) Oracle s JDK 상용계약 (2) Red Hat s IcedTea JDK 상용계약 (3) Open JDK 무상사용 (Free Open JDK) (4) Azul s Zulu JDK 상용계약 17

18 (1) Oracle JDK 상용계약검토 기술적인관점에서변경사항없음 다운로드절차만변경될뿐, 상용화적용을위한새로운변경불필요 경제성의관점에서는 (On the business side) 가장비싸고, 매우높은비용을각오해야함 (the huge, most expensive option) 18

19 (2) Red Hat IcedTea JDK 상용계약검토 Red Hat 은 2017 년 12 월이후자체제품에 Oracle Java SE 번들제공중단 기존고객도 2018 년 11 월말까지만 Oracle Java SE 적용가능 Red Hat 은 Open JDK 소스코드로부터자체제작한 JDK 로전환중. 자체 JDK 는 Open JDK 기준으로 2020 년 10 월까지만제공예정. 그러나, Red Hat 외수많은다양한시스템환경들을고려할때, 적합하지않은옵션임을확인함. 19

20 (3) 무료 Open JDK (Free OpenJDK Option) 무료 Open JDK 를사용하려는기업고객은매 6 개월마다새로운 JDK 버전으로업그레이드해야함. JDK 8 이하버전에서 9 이상으로업그레이드할경우, Java Platform Module System 으로의변경으로인해잠재적으로매우중요한검증과정요구됨. 대부분의 third-party libraries & frameworks 은아직이러한높은수준의변경에대하여준비되어있지않은상황임. 향후출시될 JDK 도이와같이, 하위버전에대한호환성등기업고객에게상당한부담을초래할수있는큰변화를포함할수있는가능성이많음. 20

21 (4) Azul Zulu JDK 상용계약검토 Azul 의상용고객은 Open JDK 소스코드기반으로표준화된 JDK 를더긴기술지원조건으로사용가능함. Azul 의모든 Java 버전에대한연장된기술지원정책은기업고객의시스템운영에있어서지속적인보안성과안정성을제공함. Azul 은 Java Community Process (JCP) 의상임멤버기업으로참여하고있으며 (the Executive Committee), Java SE 9, 10 and 11 를위한전문가그룹의멤버기업으로활동중임. 21

22 Azul Systems 소객

23 Azul s Java Background 100% Java 사업에만집중 세계최고수준의 Java 기술지원능력보유 (Mission-critical JVM deployments) 기존고객의 97% 이상이계약갱신 검증된전문성을바탕으로기술중심의 JVM 상품제공 Java Community Process(JCP) 상임임원멤버기업 (Executive Committee, OpenJDK committers) OpenJDK 6 프로젝트리더역할수행 표준화, 호환성제공 - TCK test suites 권한보유 주요 Java 버전에대한 10 년이상의기술지원정책제공 23 Copyright Azul Systems Azul Systems

24 Azul Systems 고객사 24 Powering the Most Demanding Applications of Enterprise Customers Worldwide

25 25 Azul Systems Partners

26 차별성 & 호환성검증 % open source. OpenJDK 기반으로표준화된 Java SE 호환성제공 2. Performance. 오라클도동일한 Open JDK 소스코드를사용하는것처럼, 동일한소스코드를사용함으로 Performance 는크게차이가나지않음. 3. Certified Compliance. Azul 은 OpenJDK Community 의공식 TCK License Agreement (OCTLA) 를보유하고있음. 따라서, Zulu JDK 는정당성과 Java SE 호환성이인증되어있으며, Technology Compatibility Kit (TCK) 및 14 년이상누적된자체품질검증과정으로 100,000 번이상의테스트를거친후제공됨 4. Certified Non-Contamination. 엄격한사용권한검증과정을통해추가적인권한확보나요구사항이발생하지않도록검증과정을거친후제공됨 5. Patent Indemnification. 더안전한권한을위하여 JCP Specification licenses 에서제공되는 IP 권리외, Azul 자체 IP 권한제공 26

27 차별성 & 호환성검증 1. Oracle JVM vs. Azul JVM A. 동일한소스코드로부터개발 (the same source code) B. 동일한 Java SE 호환성 C. 동일한 100,000 번이상의 TCK 테스트검증 (same 100,000 TCK tests) 2. Certification information A. Azul 은공식 TCK 테스트권한을보유하고있음 (Azul is a TCK test suite licensee) B. Azul 은제공되는소스코드가테스트되었으며, 소스코드내어떠한사용권한문제가 없음을검증 3. OpenJDK is the reference implementation for Java SE A. Java 7 이후, 모든 Java 개발은오픈소스기반하에진행됨 B. 오라클및다른상임멤버기업들의엔지니어들이 Open JDK 프로젝트를위해협력중 There is no secret source in Oracle JDK, including performance A. 오라클또한 Open JDK 소스코드를기반으로하는 JDK 제공함

28 벤치마킹테스트사례 Performance Comparison SpecJVM2008 Benchmark * Results scaled to Oracle performance Zulu Oracle HotSpot Performance differences are not significant; similar differences can occur when running the same benchmark twice on the same JVM. SpecJVM2008 Benchmark compiler.compiler compress crypto.aes crypto.rsa crypto.signverify derby mpegaudio scimark.fft.small scimark.lu.large scimark.lu.small scimark.monte_carlo scimark.sor.large scimark.sor.small scimark.sparse.large scimark.sparse.small serial startup.compiler.compiler startup.compress startup.crypto.aes startup.crypto.rsa startup.crypto.signverify startup.helloworld startup.mpegaudio startup.scimark.fft startup.scimark.lu startup.scimark.monte_carlo startup.scimark.sor startup.scimark.sparse startup.serial startup.sunflow startup.xml.transform startup.xml.verification sunflow xml.transform xml.validation geometric mean Azul Zulu 8u112 vs. Oracle Java SE 8u111 14% Worse Same 4% Better 22% Better 9% Better 6% Worse 4% Better 5% Worse Same Same 6% Worse Same Same Same 1% Worse 4% Worse 7% Worse 2% Better 6% Better 22% Better 10% Better 26% Better 3% Better 1% Worse 2% Better 1% Worse 2% Better 1% Worse 4% Worse Same Same 6% Worse 2% Worse 24% Better 9% Better 1% Better 28

29 Azul s Java Portfolio: Zing, Zulu and Zulu Embedded 29 Zing: A better runtime for Java Workloads Azul s flagship JVM focused on better metrics Removed GC as a concern for Java-based operations Applications range from web-based systems to low-latency systems Proven in on-premise servers, in containers, and in the Cloud Zulu Enterprise: Java with Superior Support Free and Open Source Java (builds of OpenJDK) Certified Java SE 9,8,7,6 fully compliant with Java standards Identical performance vs. OpenJDK and Oracle HotSpot Long support lifetime (10+ years) Timely security updates quarterly and out-of-band, as needed Zulu Embedded: Java for Embedded & IoT Custom Java SE builds for target form factors 100% open source builds of OpenJDK Certified Java SE builds Verified non-contaminating open source license Support for Linux and Windows -- Intel, ARM, PowerPC Copyright Azul Systems Azul Systems

30 Azul 상품의기술지원정책 30 Java SE Release Azul Lifecycle Java SE GA Date Java 6 LTS Dec 2006 Apr 2013 Java 7 LTS Jul 2011 Apr 2015 Java 8 LTS Mar 2014 Jan 2019 Java 9 MTS Sept 2017 Mar 2018 Java 10 Preview Mar 2018 Sept 2018 Java 11 LTS Sept 2018 Mar 2019 End of Oracle Public Updates Oracle Java SE Azul Zulu Enterprise Azul Zulu Embedded Azul Zing CURRENT OFFERINGS Dec 2018 (12 yrs) Jul 2022 (11 yrs) Mar 2025 (11 yrs) Mar 2018 (6 mo) Sept 2018 (6 mo) Dec 2019 (13 yrs) Jul 2023 (12 yrs) Mar 2026 (12 yrs) Mar 2020 (2.5 yrs) FUTURE OFFERINGS (est.) Sept 2026 Sept 2027 (8 yrs) (9 yrs) Dec 2019 (13 yrs) Jul 2023 (12 yrs) Mar 2026 (12 yrs) Community Only Apr 2017 (11 yrs) Dec 2021 (10.5 yrs) Mar 2025 (11 yrs) Community Only Community Only Community Only Community Only Sept 2027 (9 yrs) Sept 2026 (8 yrs) Java 12 Preview Mar 2019 Sept 2019 Not Supported Community Only Community Only Community Only Java 13 MTS Sept 2019 Mar 2020 Not Supported Mar 2023 (3.5 yrs) Community Only Community Only Java 14 Preview Mar 2020 Sept 2020 Not Supported Community Only Community Only Community Only Java 15 MTS Sept 2020 Mar 2021 Not Supported Mar 2023 (2.5 yrs) Community Only Community Only Java 16 Preview Mar 2021 Sept 2021 Not Supported Community Only Community Only Community Only Java 17 LTS Sept 2021 Mar 2022 Sept 2029 (8 yrs) Sept 2030 (9 yrs) Sept 2030 (9 yrs) Sept 2029 (8 yrs) Java 18 Preview Mar 2022 Sept 2022 Not Supported Community Only Community Only Community Only Java 19 MTS Sept 2022 Mar 2023 Not Supported End of Commercial Support Mar 2026 (3.5 yrs) Community Only Community Only

31 Java SE Version Java SE Lifecycle: 5+ Years (LTS) 16 Azul Zulu LTS Support 8 years active (LTS) Azul Zulu MTS Support Bridge to next LTS (LTS) 11 (LTS) (LTS) 8 (LTS) 7 (LTS) Oracle Publicly available binaries (unsupported) 2019 Oracle Commercial Support Oracle Extended Commercial Support 2022 Azul Zulu Community Builds Azul Zulu Production Commercial Support 7 (LTS) Azul Zulu Extended (Passive) Commercial Support 31

32 Migrating From Oracle JDK

33 33 Simple Migration 1. Determine JDK version for migration and download 2. Install new JDK binary 3. Modify JAVA_HOME and PATH variables 4. Test application 5. Deploy 33

34 Conclusion

35 Java Is Changing New JDK release cadence No more free Oracle JDK in production and commercial OpenJDK updates only for six-months Consider your JDK migration options Azul can help! Extended support available for more releases Easy migration path for applications Tested and supported (not just porting fixes) 35

36 별첨 1. Oracle 가격구조 Java SE Desktop Subscription Pricing Volume Subscription Metric Monthly Subscription Price Annual Subscription Price Named User Plus US$2.50 US$ ,000-2,999 Named User Plus US$2.00 US$ ,000-9,999 Named User Plus US$1.75 US$ ,000-19,000 Named User Plus US$1.50 US$ ,000-49,999 Named User Plus US$1.25 US$ ,000+ Contact for details Java SE Subscription Pricing Volume Subscription Metric Monthly Subscription Price Annual Subscription Price 1-99 Processor US$25.00 US$ Processor US$23.75 US$ Processor US$22.50 US$ Processor US$22.00 US$ ,000-2,999 Processor US$17.50 US$ ,000-9,999 Processor US$15.00 US$ ,000-19,999 Processor US$12.50 US$ ,000+ Contact for details 1 JAVA SE 데스크톱가격정보 가격예시 : 5000 명의데스크톱사용자의경우, 사용자는 3,000-9,999 명이며이는 21 달러입니다. 1 년 Java SE Desktop 구독가격은 US $ 21 * 5000 * 1 = US $ 105,000 3 년 Java SE Desktop 구독가격은 US $ 21 * 5000 * 3 = US $ 315,000 입니다. 2 JAVA SE 가격정보 ( 서버 ) 가격예시 : 8 개의코어를갖춘 1 개의 Intel Xeon 시리즈프로세서의경우 Oracle Java SE 구독가격은 US $ 25 * 12 = 연간단가 US $ 300 입니다. 300 달러 ( 단가 ) * 8 ( 코어번호 ) * 1 ( 연도번호 ) * 0.5 ( 핵심요소 ) = US $

37 Thank You

38 38

39 Azul s Flagship Java Runtime Zing A Better JVM with Better Metrics Consistent Low Latency & Pauseless Operation Slashes Java Warm-Up Time Reduces Troubleshooting Improves Business Outcomes 39 Completely Removes Java GC as a Factor Better behavior across a wide variety of workloads and use cases Pauseless Operation with JVM Heaps up to 2 TB

40 Powering Zing: 3 Key Technologies Enabled by the C4 Collector, Falcon JIT compiler, and ReadyNow! technology, Zing completely eliminates traditionally problematic Garbage Collection and Warm-up issues, along with the hiccups, stalls, freezes, jitters, and long tails that are typically associated with Java performance. C4: Azul s GC technology eliminates Java stalls, pauses and jitter Falcon: Zing s highperformance JIT compiler based on LLVM ReadyNow!: Zing runtime technology that solves Java s Warm-up problem 40

41 Zing s Falcon JIT Performance Falcon consistently surpasses C2-based Oracle HotSpot and performance of prior versions of Zing (C2 was the default Zing JIT compiler until release) SPECjvm x faster than Oracle HotSpot on crypto workloads 15-19% faster than Zing using C2 Low-latency workload (existing Zing customer, trading infrastructure): 18-24% faster than prior release of Zing 2-5% faster than Oracle HotSpot ( and HotSpot is unable to meet latency SLAs) 10% faster that Oracle HotSpot on Cassandra benchmarks using Skylake-based servers (Intel Xeon E5-xxxx) 24% faster than HotSpot running MonteCarlo Simulations 10% faster than Zing when using C2 Note: All benchmark results using JDK8 versions of Zing

42 Zing, Falcon and the analyst community By adopting LLVM for its next-generation JIT compiler, Azul is leveraging an innovative and dynamic global community that is continually upgrading performance across multiple languages and technologies. Bringing LLVM to the Zing runtime will help enterprises drive additional value from their ongoing investments in Java-based applications and infrastructure. -- John Abbott, 451 Research 42

43 Zing Drives Down CAPEX and OPEX Zing was designed to be a drop-in replacement for Oracle HotSpot No need to change your application code in order to gain benefit from Zing No need to re-architect Requires fewer servers or AWS instances to meet the same SLA Gain 2X-3X or more in additional capacity Keep the same AWS instance size and core count Eliminates the need for constant JVM tuning Fewer JVM tuning flags = fewer ways to go wrong Zing is designed to just work from startup Reduces the need for JVM performance specialists Now your best engineers can focus on building competitive new features ELA-based pricing and flexible subscriptions simplify your budgeting 43

44 Zing and Zero-GC Frameworks Zero-GC Frameworks are widely deployed throughout capital markets When rules are followed, very few GC artifacts Zing helps ensure that Zero-GC frameworks stay that way Even when developers make coding errors When 3 rd -party libraries get included Zing solves Java warm-up issues that Zero-GC frameworks can t touch Only Zing has ReadyNow technology Make sure your systems are ready at market open or other critical times Make the right compilation decisions automatically based upon prior runs Zing delivers better JIT-compiled code Falcon replaces the C2 JIT compiler in Oracle HotSpot (and older versions of Zing) More optimization PLUS better use of new Intel and AMD hardware instructions 44

45 Zing and the Cloud: Customer Use Case Video advertising company Cassandra cluster running on 6x AWS i3.2xlarge instances Approx. 80/20 write/read split Data read and written with quorum consistency 6 client machines sending requests collocated in the same Azs SLA requirements for read operations: 20ms at 99.9% 50ms at 99.99% 100ms at % (not a typo, last 9 hard to maintain on AWS) Oracle HotSpot w/g1gc can maintain ~4K TPS before SLA breach Zing: can maintain ~21K TPS before SLA breach 45

46 Proven Zing Advantages for Kafka Deployments Lower average and peak message latency Regardless of message size and read/write transaction mix Meet performance targets and SLAs Deliver better quality of service to human-facing Web applications Get diagnostic data to downstream systems quickly Reduced Engineering and Operational Effort Improved developer productivity and app time to market Less operational risk Eliminates constant JVM tuning as Kafka cluster utilization increases Improved Infrastructure Economics Better server efficiency and utilization In SLA-sensitive environments, fewer AWS instances are required when using Zing 46

47 Kafka on HotSpot with G1GC vs. Zing C4 Collector HotSpot Zing HotSpot Zing 47

48 Kafka Read-Only Workload HotSpot vs. Zing HotSpot (5K TPS) vs. Zing 5K TPS Maximum Latency in Time Interval HotSpot (5K TPS) vs. Zing 5K TPS Maximum Latency By Percentile Distribution 48

49 Zing on Amazon Cloud (AWS) Azul Zing Oracle HotSpot 49

50 Zing Benefits for Cassandra Deployments Better Raw Performance Real-time requirement demands fast execution From machine scale to human scale, Zing delivers Tomb Unmatched Reliability Eliminates Java-based disruptions, glitches and pauses Better customer experience = more revenue Meet M2M and human-scale SLAs Competitive and Time-to-Market Challenges Meet your Cassandra service delivery standards Increased Value Gain efficiency and maximize utilization Stop spending on JVM tuning Minimize missed opportunity cost Lower Capex and Opex 50

51 A Simple Visual Summary This is Cassandra on HotSpot This is Cassandra on Zing 51

52 Zing delivers value for Cassandra sites Zing reduces Cassandra compaction delays & eliminates client disconnects Zing improves uptime, reliability, and performance efficiency 52 Zing eliminates the need for tedious GC and JJVM troubleshooting and tuning

53 Zing adds value throughout the Cassandra lifecycle Evaluation & Design Pilot & Build-out Production / Expansion Reduces performance risks Ensures stability and consistency Added value when Elastic, Lucene, Solr, Spark or Storm in application stack Eliminate tedious, expensive, complex and recurring Java tuning Improves reliability by helping to eliminate client disconnections Improves scalability Maximizes overall performance efficiency Improves response time Greatly reduces peak latency, helps meet SLAs Proven to greatly improve quality of service 53

54 Azul s Real-life Cassandra Experience Customer using Cassandra with Zing Top online digital media service provider Smartphone manufacturer Large discount retailer in US Social marketing firm Credit Card Processing Firm Online gaming company Online digital media service provider Online auction site Capital market trading firm Financial services firm Major online discount travel site US cable operator Mobile marketing firm Global messaging solution provider Video advertising platform company Specific Use Case Recommendation engine Fraud detection and security Global SKU catalog HA social infrastructure CC fraud machine learning Authentication system for online game users Data storage for user personalization/profile Data storage for user profile + recommendation system Market order system Market Tick consolidator Pricing application Web based DVR / Advertisement delivery system Mobile advertisement delivery / tracking Data store / metadata archiving Digital advertising real-time marketplace (write-intensive) 54

55 Cassandra Performance: Oracle JVM vs. Zing Oracle JVM: msec stalls Zing (drawn to scale) 55

56 Some Cassandra-stress Benchmarking Results Max Latencies by Operations per Second. Cassandra setup 3 nodes, Replication Factor of 3, Quorum set to 3, across three physical servers. 56

57 Zing: Better Cassandra Performance at 4X the Workload 57

58 Zing and Cassandra: When SLAs Matter In our business speed is critical, but speed means nothing if you fail SLAs. Using the latest versions of Zing we were able to reliably maintain our SLA standards at dramatically higher speeds than we were able to previously do with the Oracle HotSpot JVM. -- Garry Turkington, CTO, Improve Digital 58

59 Zing Eliminates Java-Based Cassandra Issues Oracle HotSpot Azul Zing 59

60 Zing Powers Search for Elastic, Lucene and Solr FIXME 60 Lucene-based Search products, including Elastic and Solr, are built in Java and Scala, both of which run on a JVM JVM-based performance artifacts can impact search and sharding latency, ultimately threatening SLAs and even cluster stability With legacy Java runtimes like Oracle HotSpot, search implementations face the tradeoff of larger indices and more in-memory data structures vs. GC pauses and other JVM glitches that limit performance and availability Zing removes all practical limits on Java heap size Elastic, Lucene and Solr can take full advantage of the memory in today s servers Zing also simplifies search application engineering and support costs, and takes the need for JVM tuning off the table

61 Solr Benchmark Results Zing C4 vs. HotSpot with G1 results were conclusive In both use cases, using large (16G) and smaller (4G) heaps, Zing delivered better performance for different Solr workloads Max application pauses for Workload 1 (fulltext) were 1661ms with G1, and 67ms with Zing Max application pauses for Workload 2 (metadata) were even more dramatic G1 s worst pause was 184,684ms vs 107ms for Zing G1 performance started degrading at the 99 th percentile In all use cases Zing pause time is flat past 5 nines Results were charted using the open source jhiccup tool Conclusion: Zing s C4 collector delivers better and more consistent performance vs. HotSpot s G1 in multiple Search benchmarks 61

62 Workload #1 (Fulltext) benchmark: 32G RAM, 16G Java Heap 62

63 Metadata Search Workload -- 8GB RAM, 4GB Heap 63

64 Fulltext Search Workload: 32 GB RAM, 16 GB Heap 64

65 Zing: proven value for online retail Since 2005, Azul has been powering online retail sites like Belk, Saks 5 th Avenue and J Crew across dozens of Black Fridays and Cyber Mondays Why Azul? User experience is absolutely critical for online retail Glitches, pauses and stalls lead to unpredictable and unacceptable application performance Once customers churn away, they often move to a competitor and they don t return Worse, they share their dissatisfaction with social media, and can damage a brand within minutes Look familiar? 65

66 Creating the Best Digital Experience Drives Competitive Advantage for Online Retail of users expect a website to perform well, even during peak periods 66

67 But online buyers are often disappointed! 61 % 38 % 65 % had a poor experience had a poor experience on a retail site during the holidays had a poor experience on a travel site during a peak period on a brokerage site during a peak period 67

68 Keep Users Happy and Beat the Competition 60 xlower peak response time 23 xmore throughput...simply by Deploying on Zing 68 Source:

69 Zing Supported Platforms All Major Linux Distros: RHEL 5.2 or later and 6.0/7.0 or later Amazon Linux (5.10) SLES 11 sp1, sp2 and sp3 CentOS 5.2 or later, 6.0 or later, 7.0 or later Ubuntu LTS, and LTS Debian: Stretch, Wheezy and Jessie Oracle Linux Alpine Linux Red Hat MRG VMware, KVM, Docker Amazon AWS, Azure & Multiple Private Clouds Java Versions: 8, 7, 6 Hardware: Intel/AMD x64 69 Copyright Azul Systems Azul Systems

70 Zing summary: choose a better behaving Java As a drop-in alternative to existing JDK or JRE packages, Zing runs unmodified Java applications and infrastructure components while delivering better responsiveness, improved carrying capacity, reduced timeouts and error counts under load, and an overall smoother execution experience. Zing completely eliminates traditionally problematic Garbage Collection issues, along with hiccups, stalls, freezes, jitters, and long tails that are typically associated with Java applications. Using Zing leads to operational efficiencies in production, in Engineering, operations, and DevOps processes, in time-to-market and time-toperformance, and often in application design and performance tuning efforts. 70

71 Zulu: Our Multiplatform Builds of OpenJDK Zulu 100% Open Source Java Tested and Certified Java 9, 8, 7 and 6 Flexible Support Plans and Pricing Linux, Windows, & Mac OS X Server, Desktop, Cloud, Client, Virtualized, Juju & Docker 71 Backed By Azul s Global Support Organization Free to Download and Use Without Restriction Fixes Upstreamed to the OpenJDK Community

72 Looking for a Better Java Support Alternative? Choose Zulu Enterprise Certified, tested builds of OpenJDK Economical, world-class Java support Windows, Linux or Mac; simple subscription model 10+ year lifetime for major Java releases: You control your upgrade timing Rapid out-of-release cycle access to security updates Free downloads % Open Source Cloud and Virtualization-ready Compliant with Java SE versions 6, 7, 8 and 9 Docker Hub, Azure, AWS, Canonical Charm Store, plus Azul.com 72

73 Why Adopt Zulu Enterprise vs Oracle HotSpot? Java SE Advanced: is expensive, inflexible and unnecessary Oracle s field-ofuse licensing policies are restrictive Same code base as Oracle HotSpot same performance, 100% open source Leverage Java Community for new Embedded and IoT designs 73

74 Zulu: the Largest Commercial OpenJDK distribution 74 Zulu Platform/OS Support Linux: 32/64-bit RHEL 7, 6 and 5.9+ or later SLES 12 sp1, 11 sp1/2/3/4 CentOS 7, 6, and 5.9+ or later Ubuntu 16.04, 14.04, LTS Debian Stretch, Wheezy, Jessie Wind River Linux Oracle Linux 6 and 7 or later Alpine Linux Windows: 32/64-bit Client 10, 8.1, 8, 7, IoT/Mobile/Nano 10, Server 2016, 2012 R2, 2012, 2008 R2 Mac OS X VMware, Hyper-V, KVM Intel, ARM, PowerPC 32 and 64-bit Azure, AWS, Google, Snappy, Docker, Cloud Foundry, Juju JDK Versions: Java SE 6, 7, 8 & 9 Copyright Azul Systems Azul Systems

75 Introducing Zulu Embedded: OpenJDK for the IoT Customizable Multiple 32 and 64-bit processor architectures Customized form factors to meet the specific requirements of the target device Builds as small as 14 MB (Compact 1) available today Custom packaging or standard file types such as ZIP, MSI and.deb Java SE 6, 7 and 8; Java SE 9 EA Shipping since November 2014 Millions of devices using Zulu Embedded in the field today 75

76 Why Zulu Embedded? Oracle s Embedded Java licensing is expensive Oracle s field-ofuse licenses are restrictive Same code base as Oracle HotSpot same performance, 100% open source Leverage Java Community for new Embedded and IoT designs 76

77 Zulu Embedded: Customizable Packages Multiple OS options: Windows (client/server), all Linux distros CPUs & Bitness: 32/64 for x86, ARM, and PPC Java versions: all patch levels for Java 6, 7, 8, 9 EA Packages: ZIP, RPM, DEB, MSI Bundles: 6 unique versions JDK: 90MB JRE: 72MB JRE: 45MB CP3: 21MB CP2: 18MB CP1: 14MB 77 Headful Headless