OSv: probably the Best OS for Cloud workloads you've never heard of Roman Shaposhnik, Director of Open

Transcription

1 OSv: probably the Best OS for Cloud workloads you've never heard of Roman Shaposhnik, Director of Open

2 # whoami Director of Open ASF junkie Member, VP of Incubator Co-founder of Apache Bigtop Used to work for Sun micro around Solaris Used to be a kernel hacker Linux, Plan9 11/17/14 2

3 Why am I talking about OSv The most exciting development in kernel/ OS space in a long time How distributed systems and μservices were meant to be deployed A non-doa way to run JVM on bare metal A very exciting building block for PaaS 11/17/14 3

4 Why am I talking about OSv The most exciting development in kernel/ OS space in a long time How distributed systems and μservices were meant to be deployed A non-doa way to run JVM on bare metal A very exciting building block for PaaS Expecting a beer from Cloudius Systems 11/17/14 4

5 Distributed system architecture in 30s Zookeeper I am alive Who Am I? What do I do? HBase region server HBase region server HDFS 11/17/14 5

6 Failure scenario Zookeeper Node died YOU! Pitch in! Aye, aye captn HBase region server HDFS 11/17/14 6

7 Failure recovery Zookeeper I am alive Who Am I? What do I do? HBase region server HBase region server HDFS 11/17/14 7

8 Zooming in on each node service #1 service #N 11/17/14 8

9 Anatomy of a μservice μservice code [Java] Virtual Machine Stuff Hardware 11/17/14 9

10 How are we doing it today? Puppet, Chef μservice code Linux kernel pkg1 [Java] Virtual Machine. pkgn OS Hardware Huge VM image

11 Is there a better way? Applicacon- specific stacc linking μservice code [Java] Virtual Machine libfs, libc, libjvm Tiny VM image AKA unikernel vhardware Hardware- assisted virtualizacon Hardware

12 What the heck is a FOOkernel? What OS design courses have taught us? microkernels vs. monolithic kernels What did they left behind? exokernels, nano What they should ve taught us instead: unikernels, anykernels

13 Unikernels Unikernels: library operating systems for the cloud came out in 2013 A library operating system A kernel that can only support one process

14 Anykernels Programming discipline for kernel code reuse The Design and Implementation of the Anykernel and Rump Kernels by A. Kantee Capabilities NetBSD filesystems as Linux processes User-space TCP/IP stack

15 OSv from Cloudius Systems A unikernel for POSIX and memory managed platforms (JVM, Go, Lua) Anykernel ish E.g. ZFS Runs on top of KVM, Xen, VirtualBox, VMWare Looks like an app to the host OS Small, fast and easy to manage at scale

16 OSv manifesto Run existing Linux applications Run existing Linux applications faster Make boot time ~= exec time Explore APIs beyond POSIX Leverage memory managed platforms (JVM, Go) Stay open

17 What s inside? C++ kernel code ZFS TCP/IP virco disk ifconfig.so libjvm.so iface dynamic linker kernel threads user threads single address space in kernel mode

18 Anything it can t do? A 100% replacement for a Linux kernel No fork()ing No process isolation The least amount of device drivers ever

19 Virtualization vs. performance Network-intensive apps: unmodified: 25% gain in throughput# 47% decrease in latency non-posix APIs use for Memcached: Compute-intensive apps: YMMV 290% increase in performance

20 Van Jacabson s net channels send/recv send/recv socket lock socket socket lock channel TCP lock TCP TCP IP lock IP IP classifier iface iface iface iface app thread kernel (IRQ) app thread kernel (IRQ) Tradiconal TCP/IP stack OSv TCP/IP stack

21 Memory management in UNIX JVM Heap JVM Heap Process Memory Process Memory OS Memory

22 Memory management in OSv JVM Heap Process Memory OS Memory

23 JVM balooning (no more -Xmx) OS object JVM Heap

24 Turbo charging JVM GC object 1 object 2

25 Turbo charging JVM GC object 1 object 2

26 Turbo charging JVM GC object 1 object 2 CPU MMU assisted tracking table

27 Apache Cassandra The good: mostly working tons of potential performance improvements JVM balooning could do very nicely The bad: performance on-par with bare metal The ugly: OSv s mmap() pecularities 11/17/14 27

28 Apache Zookeeper The good: mostly working no modifications: Bigtop rpm -> OSv image The bad: no cycles to benchmark/validate The ugly: no canonical image (Bigtop to the rescue!) 11/17/14 28

29 Apache Hadoop The good: HDFS seems to be working The bad: not sure what to do abot YARN no cycles to benchmark/validate The ugly: need to patch Hadoop common 11/17/14 29

30 Didn t I tell you no forking? From org.apache.hadoop.fs.df: public String getfilesystem() throws IOException { - if (Shell.WINDOWS) { + // if (Shell.WINDOWS) { this.filesystem = dirfile.getcanonicalpath().substring(0, 2); return this.filesystem; - } else { - run(); - return filesystem; - } + // } else { + // run(); + // return filesystem; + // } } 11/17/14 30

31 Apache HBase Next thing on the list short of hiring an intern, not sure when it is happening First attempt at general Bigtop-based OSv packaging A case of 100% stateless application A good companion to Cassandra work 11/17/14 31

32 Bigtop s perspective Apache Bigtop is to Hadoop what Debian is to Linux Linux Packaging/Integration init.d hooks (start/stop/restart) OSv Packaging/Integration no way to prepare a java command line 11/17/14 32

33 Bigtop <3 Docker Assume 100% Docker integration OSv as Docker accelerator Universal containers IaaS players entering the market Joyent s Smart Data Center (now Open Source!) 11/17/14 33

34 But what about Docker? Applicacon- specific stacc linking μservice code [Java] Virtual Machine libfs, libc, libjvm Docker image Jailed FS, net, etc. Common, shared kernel Hardware

35 Docker!= LXC $ docker run roman/petclinic Socially-driven image sharing Canonical Ubuntu Pivotal Roman Tomcat PetClinic Google service #1 $ docker run PetClinic service #N Go

36 Bigtop as a universal packaging platform What is a package? a [partial] image of a filesystem What is a docker container? an image of a filesystem What is an OSv image? an image of a filesystem (on-disk ZFS) 11/19/14 36

37 Why should it work this time? Unikernels/exokernels back in 90 JVM-on-bare-metal (Azul, BEA, etc.) back in 00 Things they didn t have back then HW-assisted virtualization (KVM, XEN, etc.) Elastic infrastructure oriented architectures CloudFoundry

38 Elastic, next generation datacenter Commodity, rack-provisioned Hardware Commodity, JeOS to get to Docker++ CoreOS, SmartOS Docker++ as a common backed OSv (really KVM, XEN) GitHub for μservies images

39 Finally killing DevOps Ops (IT) maintains the bare OS Devs maintain the images

40 Finally killing DevOps Ops (IT) maintains the bare OS Devs maintain the images

41 Questions? Imagine no platorms I wonder if you can No need for PAAS or IAAS A brotherhood of bare metal Imagine there is no VM It's easy if you try No host below us Above us only apps

42 Thanks!