FreeBSD 7 & diablo

Hi,

With FreeBSD just around the corner, have anyone tested
diablo on it? Any 
improvement from 6-STABLE? I still get the most out of my
feeders that 
runs on FreeBSD 4-STABLE.

/Bjorn
_______________________________________________
Diablo-users mailing list
Diablo-usersopenusenet.org
htt
p://www.plig.net/mailman/listinfo/diablo-users

One other question.

Is anyone running Diablo on DragonFly BSD? And how is that
in compare to 
FreeBSD?

/Bjorn

On Tue, 19 Feb 2008, Bjorn Hellqvist wrote:

>
> Hi,
>
> With FreeBSD just around the corner, have anyone tested
diablo on it? Any
> improvement from 6-STABLE? I still get the most out of
my feeders that
> runs on FreeBSD 4-STABLE.
>
> /Bjorn
> _______________________________________________
> Diablo-users mailing list
> Diablo-usersopenusenet.org
> htt
p://www.plig.net/mailman/listinfo/diablo-users
>
_______________________________________________
Diablo-users mailing list
Diablo-usersopenusenet.org
htt
p://www.plig.net/mailman/listinfo/diablo-users

Le (On) Tue, Feb 19, 2008 at 10:50:24AM +0100, Bjorn
Hellqvist ecrivit (wrote):
> 
> Hi,
> 
> With FreeBSD just around the corner, have anyone tested
diablo on it? Any 
> improvement from 6-STABLE? I still get the most out of
my feeders that 
> runs on FreeBSD 4-STABLE.

There are two features I wanted to use on some FreeBSD
servers: the real LACP
implementation (lagg), and tmpfs.

Tmpfs is only found on 7.x, but lagg has been backported to
6-STABLE AFAIR.

Seems fine atm, I store history, queue, and spool files in
RAM, and I needed a
smoother behavior than the casual md memory disks.

Depending on the load, the hardware, and your objectives you
might want to give
it a try.

Cheers,

-- 
Sameh Ghane
_______________________________________________
Diablo-users mailing list
Diablo-usersopenusenet.org
htt
p://www.plig.net/mailman/listinfo/diablo-users

On Tue, 25 Mar 2008, Sameh Ghane wrote:

> Le (On) Tue, Feb 19, 2008 at 10:50:24AM +0100, Bjorn
Hellqvist ecrivit (wrote):
>>
>> Hi,
>>
>> With FreeBSD just around the corner, have anyone
tested diablo on it? Any
>> improvement from 6-STABLE? I still get the most out
of my feeders that
>> runs on FreeBSD 4-STABLE.
>
> There are two features I wanted to use on some FreeBSD
servers: the real LACP
> implementation (lagg), and tmpfs.
>
> Tmpfs is only found on 7.x, but lagg has been
backported to 6-STABLE AFAIR.
>
> Seems fine atm, I store history, queue, and spool files
in RAM, and I needed a
> smoother behavior than the casual md memory disks.
>
> Depending on the load, the hardware, and your
objectives you might want to give
> it a try.
>

And you have not found any stability issues?

/Bjorn
_______________________________________________
Diablo-users mailing list
Diablo-usersopenusenet.org
htt
p://www.plig.net/mailman/listinfo/diablo-users

Le (On) Wed, Mar 26, 2008 at 01:24:54PM +0100, Bjorn
Hellqvist ecrivit (wrote):
> 
> > Tmpfs is only found on 7.x, but lagg has been
backported to 6-STABLE AFAIR.
> >
> > Seems fine atm, I store history, queue, and spool
files in RAM, and I needed a
> > smoother behavior than the casual md memory
disks.
> >
> > Depending on the load, the hardware, and your
objectives you might want to give
> > it a try.
> 
> And you have not found any stability issues?

None at the moment with diablo (feeder only).

As often with FreeBSD's HEAD branch, it is very stable for
me, though the code
is recent.

I often use -CURRENT on production servers actually 

And as some seemed interested (at least Joe) about why
upgrading from older
releases for tmpfs (instead of md(4) disks), here is why I
was waiting for it:

tmpfs is present in Solaris for years, and it is much more
easy and efficient to
manage than FreeBSD's (old) md driver.

o no fixed-size maximum memory usage:
  This means you get everything you can from RAM + swap,
whereas memory-disks
  only allocates a fixed part of memory.
  If you want to make it grow, you just have to add more
swap. md disks need to
  be destroyed and recreated AFAIK.
  It is also convenient for handling a burst of articles (I
see no « No space
  left on device » messages anymore .

o no underlying file-system to create and... tune (turn off
softupdates, atime
  updates and play with inodes mostly)

o Better memory usage:
  You only use memory for the data present on the partition.
Not for the whole
  filesystem.
  Also, where I had a lot of performance issues when asking
for a memory-disk
  larger than RAM (actually it kind of stalled when I
benchmarked it, whatever
  the memory disk type) asking for a md larger than RAM, I
have no problem
  when tmpfs hits the swap: it is still comparable to disk
performance.


I did not compare performance between tmpfs & md because
features were already
convincing me, but it *should be* and it *seems* faster. The
smaller memory
footprint also helps relieving the pain for user-land
processes.

I think it's a must for people using md disks for feed
spools or history: it
scales much better.


Doing some simple benchs won't be difficult, but as long as
the news flow fine,
I'm lazy...


I will tell you how much I can climb into the top1000 with
tmpfs instead of md


Cheers,

-- 
Sameh Ghane
_______________________________________________
Diablo-users mailing list
Diablo-usersopenusenet.org
htt
p://www.plig.net/mailman/listinfo/diablo-users