49
Making clouds: turning OpenNebula into a product Carlo Daffara, CloudWeavers

Making Clouds: Turning OpenNebula into a Product

  • Upload
    netways

  • View
    2.381

  • Download
    1

Embed Size (px)

DESCRIPTION

What does it takes to bring innovations like private clouds to small and medium enterprises? In the course of this talk we will present our experience in creating a self-service toolkit for creating a complete virtualization and cloud platform based on OpenNebula, as well as our experience gathered in tens of installations of all sizes. From scalable storage (with benchmarks!) to autonomic optimization, we will present what in our view is needed to bring private clouds to everyone, what components and additions we created to better solve our customers’ problems (from replacing industrial control systems to medium scale virtual desktop infrastructures), and why OpenNebula has been chosen over other competing cloud toolkits.

Citation preview

Page 1: Making Clouds: Turning OpenNebula into a Product

Making clouds:turning OpenNebula into a product

Carlo Daffara, CloudWeavers

Page 2: Making Clouds: Turning OpenNebula into a Product
Page 3: Making Clouds: Turning OpenNebula into a Product
Page 4: Making Clouds: Turning OpenNebula into a Product
Page 5: Making Clouds: Turning OpenNebula into a Product

The Cloud extends the traditional advantages of “ ”virtualization platforms to scale-out software. It is an opportunity to rethink hardware deployment ...

…it does not means that all clouds need to be public

Page 6: Making Clouds: Turning OpenNebula into a Product
Page 7: Making Clouds: Turning OpenNebula into a Product
Page 8: Making Clouds: Turning OpenNebula into a Product
Page 9: Making Clouds: Turning OpenNebula into a Product

http: blippex.github.io updates 2013 09 23 why-we-moved-away-from-aws.html// / / / / /

Page 10: Making Clouds: Turning OpenNebula into a Product
Page 11: Making Clouds: Turning OpenNebula into a Product
Page 12: Making Clouds: Turning OpenNebula into a Product

Why OpenNebula?

Page 13: Making Clouds: Turning OpenNebula into a Product

Four major platforms:•CloudStack•Eucalyptus•OpenNebula•OpenStack

Lots of minor ones:Proxmox, Ovirt, Ganeti, OpenQRM, Enomaly, Archipel...

Page 14: Making Clouds: Turning OpenNebula into a Product

Let's debate first about what architecture we are aiming for. OpenStack & Eucalyptus are modeled and structured around AWS EC2, that is the clear market leader in /the public cloud market.

Page 15: Making Clouds: Turning OpenNebula into a Product
Page 16: Making Clouds: Turning OpenNebula into a Product
Page 17: Making Clouds: Turning OpenNebula into a Product
Page 18: Making Clouds: Turning OpenNebula into a Product
Page 19: Making Clouds: Turning OpenNebula into a Product
Page 20: Making Clouds: Turning OpenNebula into a Product

Service Manager database 8-core 2.66 gigahertz (GHz) CPU 8 gigabytes (GB) of RAM 80 GB of available disk space RAID Level 1 or Level 10 drive*

Service Manager management server 4-Core 2.66 GHz CPU 8 GB of RAM 10 GB of available disk space

Service Manager console 2-core 2.0 GHz CPU 4 GB of RAM 10 GB of available disk space

Data warehouse management server 4-Core 2.66 GHz CPU 8 GB of RAM 10 GB of available disk space

Data warehouse databases 8-core 2.66 GHz CPU 8 GB of RAM 400 GB of available disk space RAID Level 1 or Level (1+0) drive

Self-Service Portal: Web Content Server with SharePoint Web Parts

8-Core 2.66 GHz CPU 8-core, 16GB RAM, 80 GB of available hard disk space

http: technet.microsoft.com en-us library hh524328.aspx// / / /

Page 21: Making Clouds: Turning OpenNebula into a Product
Page 22: Making Clouds: Turning OpenNebula into a Product
Page 23: Making Clouds: Turning OpenNebula into a Product
Page 24: Making Clouds: Turning OpenNebula into a Product
Page 25: Making Clouds: Turning OpenNebula into a Product
Page 26: Making Clouds: Turning OpenNebula into a Product

CloudStack 3.22MLoc

Eucalyptus 1.43MLoc

OpenStack 1.67MLoc

OpenNebula 190KLoc

Page 27: Making Clouds: Turning OpenNebula into a Product
Page 28: Making Clouds: Turning OpenNebula into a Product
Page 29: Making Clouds: Turning OpenNebula into a Product
Page 30: Making Clouds: Turning OpenNebula into a Product

I O Access pattern/ I O Characteristics/ Typical Applications

Streaming Reads 100 Reads; Large contiguous requests; 1-64 %concurrent requests. May be threaded.

Media Servers ( ideo on-demand, etc.). Virtual Tape Libraries ( TL), Application V V

Servers

Streaming Writes 100 Writes; Large contiguous requests; 1-64 %concurrent requests. May be threaded.

Media Capture, TL, Medical Imaging, VArchiving, Backup, ideo Surveillance, VReference Data

OLTP Typically 2KB 16KB request sizes; Read modify, –write, verify operations resulting in 2 reads for every write; Primarily random accesses. Large number of concurrent requests. When running SQL statements in parallel, Database will perform typically large random I Os./

Databases (SAP, Oracle, SQL), Online Transaction Servers

File Server Moderate distribution of request sizes from 4KB to 64KB, however 4KB and 64KB comprise 70 of %requests; Primarily random; Generally 4 reads for every write operation. Large number of concurrent requests during peak operational periods.

File and Printer Servers, e-mail (Exchange, Notes), Decision Support Systems

Web Server Wide distribution of request sizes from 512 bytes to 512KB; Primarily random accesses; Large number of concurrent requests during peak operational periods

Web Services, Blogs, RSS Feeds, Shopping Carts, Search Engines, Storage Services

Workstations Primarily small to medium request sizes; 80 %sequential and 20 random; Generally 4 reads for %every write operation. 1-4 concurrent requests.

Business Productivity, Scientific Engineering /Applications

http: gerardnico.com wiki data storage input output access pattern// / / _ / _ _ _

Page 31: Making Clouds: Turning OpenNebula into a Product
Page 32: Making Clouds: Turning OpenNebula into a Product

Some additional tips for Windows:

• K M badly interacts with Windows 7 8 clock drivers, causing high cpu V /consumption. Use:bcdedit /set {default} useplatformclock trueand RAW = [ type= "KVM", data = "<clock offset='localtime'/>" ]

• “Default cirrus GA driver unsuitable for Windows due to pathological Vperformance issues in text mode”This is not a problem in latest K M, however having vga std is better V =anyway

• For Windows servers with lots of small files:fsutil behaviour set disablelastaccess 1(equivalent to noatime on linux)fsutil behaviour set mftzone 2(adds another 12.5 disk space to the MFT zone)%

• Disable 8.3 namespace creation:fsutil.exe behavior set disable8dot3 1(huge boost for dirs, files with matching initial names)

Page 33: Making Clouds: Turning OpenNebula into a Product

Some useful additions:GateOne (https: github.com liftoff GateOne// / / ) a very efficient web SSH interface, similar to vncproxy. Works well added as a link in Sunstone.

Page 34: Making Clouds: Turning OpenNebula into a Product

M ray (https: github.com coriolis vmxray): a pure javascript M V X // / / Vexplorer, allows to enter inside of an image, view and extract files.

Page 35: Making Clouds: Turning OpenNebula into a Product

Some additional tools:

LibGuestFS (http: libguestfs.org// /) the swiss army knife of M images. Allows Vfor copy in and out of a M, file operations, mount guest filesystem on host, VWindows registry editing...

S3QL (http: code.google.com p s3ql// / / /) a FUSE filesystem that mounts and operates on Amazon S3, Google storage and OpenStack-compatible storage. Supports compression, encryption and deduplication. Quite useful if you want to move images to-from Amazon or Google.

IPOP (http: www.grid-appliance.org wiki index.php IPOP// / / / ) IPOP (IP over P2P) is an open-source research development project that provides a decentralized, /distributed overlay virtual network. It allows users to easily deploy PNs Vacross multiple domains, and is used in applications including virtual private clusters for cloud grid computing./

Only thin features added directly to the platform...“ ”

Page 36: Making Clouds: Turning OpenNebula into a Product
Page 37: Making Clouds: Turning OpenNebula into a Product
Page 38: Making Clouds: Turning OpenNebula into a Product

We use Ms to provide additional features:V

Guacamole (http: sourceforge.net projects guacamole// / / /) exceptional gateway that turns RDP into HTML5. Allows for pure web-based access to Windows desktops from everywhere:

Other examples: Zenoss for agentless monitoring (or Hyperic for agent-based monitors), backup tools like OpenDedup...

Page 39: Making Clouds: Turning OpenNebula into a Product

A special note on DI:V

Jerry Chen of Mware is generally credited with coining the term DI which V “V ”led to the redefinition of an entire industry. Mware had 10,400 employees. VOf that group, about one-third of them use DI-based virtual desktops as their Vprimary production work desktops. About 1,000 of them use a thin-client device as their only corporate-issued client device.

First problem: Microsoft licensing. “dedicating an entire Remote Desktop virtual server to just one single user is cheaper than most DI scenarios and gives users the same benefits, namely Vhaving their own Desktop M where they can do what they want. This single ‘ ’ V ‘user terminal server scenario is with about 475 per user (6 years) not only ’ – $60 cheaper, but less complicated and legally more safe than standard MS %

DI licensing. Decoding the Microsoft DI Licensing Arcanum, Login V " VConsulting

Page 40: Making Clouds: Turning OpenNebula into a Product
Page 41: Making Clouds: Turning OpenNebula into a Product
Page 42: Making Clouds: Turning OpenNebula into a Product

A useful trick: on hook RUNNING increase the readahead of the rotative “ ”medium (blockdev --setra 2048 or higher); you can reduce the boot time of a Windows7 machine by 35 . You time it to return to much lower values %after a predetermined interval.

Page 43: Making Clouds: Turning OpenNebula into a Product

Apps are moving to be pure webapps, no client installation needed so “ ” –you can skip Windows. Having the M ready to use also facilitates the Vadoption of new tools, and lots of experimentation. Locally hosted Ms that Vexport a web application are becoming the norm, and you can remote the remaining Windows apps through RDP, so you are actually free to build your desktop as you like. Royal Academy of Engineering, The future of computing , 2012“ ”

Page 44: Making Clouds: Turning OpenNebula into a Product
Page 45: Making Clouds: Turning OpenNebula into a Product

We have our own thin desktop based on Linux called EveryDesk (http://sourceforge.net/projects/everydesk/) that requires just 90MB of ram per desktop, or for the more daring users:

Page 46: Making Clouds: Turning OpenNebula into a Product

..you can use ChromeOS from Google. There is an unofficial build for 86 Xand x86 64 available daily here (_ http://chromium.arnoldthebat.co.uk/) that runs perfectly within OpenNebula. You create the images as non-persistent, and destroy them later on.Bt using a WB cache and snapshots to create the non-persistent images you end up with most of the pages already in cache, so boot is nearly instantaneous. Write are small and sparse, so you are basically CPU-bound and can run reasonably well 10-15 ChromeOS desktops per core.

Page 47: Making Clouds: Turning OpenNebula into a Product

38 , 5W, no moving part. Use Android aSPICE as a client+Autostart$

Page 48: Making Clouds: Turning OpenNebula into a Product

The MTBF of a standard PC is 30,000 hours or 3.4 years; for every 10c temperature increase, MTBF is cut in half. Since in many pc closet the “ ”temperature is 10 to 20 degrees higher, we assume a MTBF of 1yr.

4pc cloud: 4 failures yr, Ms on failed node are down for 5-10 mins - 20 / V >to 40 min year, 99.99 assuming available power/ %

Availability per day per month per year

99.999% 00:00:00.4 00:00:26 00:05:15

99.99% 00:00:08 00:04:22 00:52:35

99.9% 00:01:26 00:43:49 08:45:56

99% 00:14:23 07:18:17 87:39:29

Page 49: Making Clouds: Turning OpenNebula into a Product

Thanks!

Carlo [email protected]

Twitter: cdaffara