DaemonForums  

Go Back   DaemonForums > Miscellaneous > General software and network

General software and network General OS-independent software and network questions, X11, MTA, routing, etc.

Reply
 
Thread Tools Display Modes
  #1   (View Single Post)  
Old 23rd January 2013
pawaan pawaan is offline
Port Guard
 
Join Date: Jan 2013
Posts: 36
Thanked 0 Times in 0 Posts
Default what's behind scalability ?

I hope someone would help me learn about those determining factors that influence an OS scalability , more precisely the BSDs.

Million Thanks
Reply With Quote
  #2   (View Single Post)  
Old 23rd January 2013
ocicat ocicat is offline
Administrator
 
Join Date: Apr 2008
Posts: 2,876
Thanked 190 Times in 160 Posts
Default

Quote:
Originally Posted by pawaan View Post
I hope someone would help me learn about those determining factors that influence an OS scalability , more precisely the BSDs.
This is a broad topic. Can you be more specific? Are you most interested in network throughput, database sizes, filesystem sizes, realtime uses, or something else altogether? Specificity will help frame discussion.
Reply With Quote
  #3   (View Single Post)  
Old 23rd January 2013
pawaan pawaan is offline
Port Guard
 
Join Date: Jan 2013
Posts: 36
Thanked 0 Times in 0 Posts
Default

I mean network scalability

What makes a OS unscalable or more likely to have a scalability issue?
what prevents a OS from being scalable enough toward changes ?

Thanks

Last edited by pawaan; 23rd January 2013 at 02:53 PM.
Reply With Quote
  #4   (View Single Post)  
Old 23rd January 2013
ocicat ocicat is offline
Administrator
 
Join Date: Apr 2008
Posts: 2,876
Thanked 190 Times in 160 Posts
Default

Quote:
Originally Posted by pawaan View Post
I mean network scalability
Here are a couple of factors:
  • The feature set of the NIC's hardware, & the matching software driver both determine basic throughput. Not all hardware is alike, & not all driver implementations exploit the underlying chips to their fullest potential. If you are looking for the "best" performance, study the archives of the project mailing lists to find the developer's opinions.
  • Secondly, other optimizations to the network stack fall under the purview of the operating system itself. Issues such as how network packets are traced through kernel code influence performance.
For most desktop-level hardware, NIC choice & driver issues are critical. Operating system design choices don't typically enter the picture until gigabit NIC's are involved (ie. enterprise level).
Reply With Quote
  #5   (View Single Post)  
Old 23rd January 2013
pawaan pawaan is offline
Port Guard
 
Join Date: Jan 2013
Posts: 36
Thanked 0 Times in 0 Posts
Default

Thanks ocicat.what you penned helps me figure out about the issue and I hope you'll expand a bit this part:
Quote:
Operating system design choices don't typically enter the picture until gigabit NIC's are involved
No OS war is meant here , but some would argue that NetBSD lately becomes much more scalable than OpenBSD , despite their similarity in design and implementation if I am correct.
If I am to judge which is most scalable among BSD derivatives , how should I proceed to achieve an objective conclusion ?
I'm looking forward to more enlightenment.
So thankful to you.
Reply With Quote
  #6   (View Single Post)  
Old 23rd January 2013
jggimi's Avatar
jggimi jggimi is offline
More noise than signal
 
Join Date: May 2008
Location: USA
Posts: 3,645
Thanked 214 Times in 189 Posts
Default

Quote:
...how should I proceed to achieve an objective conclusion ?
By testing each BSD, on your intended hardware, with your intended network load. The term used to describe this test is benchmark.

Benchmark testing is easy. Obtaining valid results is much more difficult. Be very cautious with results obtained from any synthetic network load induced with software tools. Deploying your BSD systems on your production network would be better.

A synthetic network load must model your real load closely. Not just number of packets over time and their size, but packet direction and routing headers must also be considered. See http://www.daemonforums.org/showthread.php?t=7618 for an example of a benchmark using a synthetic network load which does NOT closely model any real network.

Last edited by jggimi; 23rd January 2013 at 05:49 PM. Reason: clarity; typo
Reply With Quote
  #7   (View Single Post)  
Old 23rd January 2013
ocicat ocicat is offline
Administrator
 
Join Date: Apr 2008
Posts: 2,876
Thanked 190 Times in 160 Posts
Default

Quote:
Originally Posted by pawaan View Post
...I hope you'll expand a bit this part:
Quote:
Operating system design choices don't typically enter the picture until gigabit NIC's are involved.
For the most part, much of the development of the *BSD family is done on desktop-quality hardware. Here, network throughput is mostly limited by the efficiency of the hardware/driver combination. In general, the quality of the network stacks are sufficient to handle the network loads placed upon them. Much of the differences seen at the enterprise level (large volumes of traffic...) are due to the fact that developers do not have access to such expensive equipment.

It should also be understood that tweaking/optimizing the network stack as implemented by the kernel is a much harder problem to solve. In comparison, writing driver code is more straight-forward. Not that either is simple, but the range of problems needing to be addressed by the kernel is far more broad.
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
PostgreSQL 9.2 beta improves scalability, adds JSON J65nko News 0 15th May 2012 09:13 PM


All times are GMT. The time now is 11:20 AM.


Powered by vBulletin® Version 3.8.4
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content copyright © 2007-2010, the authors
Daemon image copyright ©1988, Marshall Kirk McKusick