Bug 21238

Summary: poor performance; missed opportunities
Product: Documentation Reporter: stan <stan>
Component: AdvocacyAssignee: freebsd-advocacy (Nobody) <advocacy>
Status: Closed FIXED    
Severity: Affects Only Me    
Priority: Normal    
Version: Latest   
Hardware: Any   
OS: Any   

Description stan 2000-09-13 01:20:01 UTC
After an outside consultant advocated we "upgrade" from Linux to FreeBSD
our sysadmin staff invested much time and effort into evaluating FreeBSD.
A new production server was built using FreeBSD 4.0 and comparted with
Linux, all on the same hardware.  The result was a comparison of the
two OSs on the same hardware running the exact same application tests.
This type of testing is as fair as it gets.

The reasons for staying with Linux follow.

o) Linux (2.2.16 kernel) benchmarked far better in terms of our search
   code (3x faster on the same box), and slightly better for Apache
   performance.

o) The final straw was that the GDBM libraries for FreeBSD were very
   slow.

If performance of these two important tests was close to the same,
we would have switched to FreeBSD.   Instead, for now the idea
of switching to FreeBSD has been dropped and not likley to be
re-examined anytime soon.

Until fairly unsophisticated applications perform as well on
FreeBSD, these rare opportunities to convert a predominately
Linux shop to FreeBSD are a waste of time.

Fix: 

Regretably the fix is to use Linux.
How-To-Repeat: Properly benchmark GDBM and Apache based applications
using the same hardware.   Compare results between FreeBSD
and Linux.
Comment 1 bill fumerola freebsd_committer freebsd_triage 2000-09-13 02:18:48 UTC
State Changed
From-To: open->closed

User fails to provide benchmarks, specifics, details, or 
anything that would prove that he is not either clueless or a troll.
Comment 2 mark.blackman 2000-09-13 11:51:17 UTC
The problem seen seems like it might be related to a point raised by Richard
Wendland (of Netcraft) on the freebsd-current mailing list on March 22, 2000.
(http://www.FreeBSD.org/cgi/getmsg.cgi?fetch=575728+585714+/usr/local/www/db/text/2000/freebsd-current/20000326.freebsd-current)

It boiled down to the async vs. sync nature of the default file system
mounting.  These DB files tend to have lots of random I/O and "This issue
appears to be related to the traditional BSD behaviour of immediately
scheduling full disc block writes" according to Richard.

Set your FreeBSD file-system to async and see what that does for your
performance.  Linux systems (dangerously) mount by default async. FreeBSD does
not take such risks with filesystems content.  Alternately, set your
Linux filesystem to synchronous and see what happens to your Linux
performance.

Finally, a more helpful procedure might have been
a) post to freebsd-questions.
  "why is there such a performance differrence between FreeBSD and Linux
   for application X?"
b) failing that, freebsd-advocacy and/or freebsd-hackers.

I'm not sure this is the same problem but its a good starting
point.  Didn't you ask your consultant why there was such a difference?

- Mark

On Tue, Sep 12, 2000 at 05:16:49PM -0700, stan@craigslist.org wrote:
> 
> >Number:         21238
> >Category:       advocacy
> >Synopsis:       poor performance; missed opportunities
> >Confidential:   no
> >Severity:       critical
> >Priority:       high
> >Responsible:    freebsd-advocacy
> >State:          open
> >Quarter:        
> >Keywords:       
> >Date-Required:
> >Class:          sw-bug
> >Submitter-Id:   current-users
> >Arrival-Date:   Tue Sep 12 17:20:01 PDT 2000
> >Closed-Date:
> >Last-Modified:
> >Originator:     Stan Osborne
> >Release:        4.0
> >Organization:
> Craigslist
> >Environment:
> no longer available
> >Description:
> After an outside consultant advocated we "upgrade" from Linux to FreeBSD
> our sysadmin staff invested much time and effort into evaluating FreeBSD.
> A new production server was built using FreeBSD 4.0 and comparted with
> Linux, all on the same hardware.  The result was a comparison of the
> two OSs on the same hardware running the exact same application tests.
> This type of testing is as fair as it gets.
> 
> The reasons for staying with Linux follow.
> 
> o) Linux (2.2.16 kernel) benchmarked far better in terms of our search
>    code (3x faster on the same box), and slightly better for Apache
>    performance.
> 
> o) The final straw was that the GDBM libraries for FreeBSD were very
>    slow.
> 
> If performance of these two important tests was close to the same,
> we would have switched to FreeBSD.   Instead, for now the idea
> of switching to FreeBSD has been dropped and not likley to be
> re-examined anytime soon.
> 
> Until fairly unsophisticated applications perform as well on
> FreeBSD, these rare opportunities to convert a predominately
> Linux shop to FreeBSD are a waste of time.
> >How-To-Repeat:
> Properly benchmark GDBM and Apache based applications
> using the same hardware.   Compare results between FreeBSD
> and Linux.
> >Fix:
> Regretably the fix is to use Linux.
> 
> >Release-Note:
> >Audit-Trail:
> >Unformatted:
> 
> 
> To Unsubscribe: send mail to majordomo@FreeBSD.org
> with "unsubscribe freebsd-advocacy" in the body of the message
>