Re: [CentOS] Samba "use sendfile" configuration option set do disabled as default - why?
Hi
I am not saying that this is the solution but I am saying that it might be worth a short, even if it works for you I would still fine tune it depending on your configs, a lot of the time fine tuning samba is a pain and most of the time it actually comes down to crappy network performance I.E the nic seems to be working but it actually has a fault.
Regards Per Qvindesland --- Original message follows --- SUBJECT: Re: [CentOS] Samba "use sendfile" configuration option set do disabled as default - why? FROM: Christoph Maser TO: "CentOS mailing list" DATE: 24-08-2009 13:27
Am Montag, den 24.08.2009, 13:24 +0200 schrieb Per Qvindesland:
socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536
These options are often found if you search for samba tuning. Did someone actaully benchmark the results? Shouldn't tuning buffers manuall be oboslete nowadays? I somewhat doubt this is THE solution for slow sambas.
Chris
financial.com AG
Munich head office/Hauptsitz München: Maria-Probst-Str. 19 | 80939 München | Germany Frankfurt branch office/Niederlassung Frankfurt: Messeturm | Friedrich-Ebert-Anlage 49 | 60327 Frankfurt | Germany Management board/Vorstand: Dr. Steffen Boehnert (CEO/Vorsitzender) | Dr. Alexis Eisenhofer | Dr. Yann Samson | Matthias Wiederwach Supervisory board/Aufsichtsrat: Dr. Dr. Ernst zur Linden (chairman/Vorsitzender) Register court/Handelsregister: Munich – HRB 128 972 | Sales tax ID number/St.Nr.: DE205 370 553 _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Thank you for replies,
with option "use sendfile = yes" it works great (with good speed), but I am asking why is this option in CentOS disabled as default, even in standard samba build it should be (according to Samba release notes) enabled as default.
So my question is, why in CentOS´s Samba build is this option as default disabled. Is it experimental? Is it potentially dangerous? (Something like oplocks?)
2009/8/24 Per Qvindesland per@norhex.com:
Hi
I am not saying that this is the solution but I am saying that it might be worth a short, even if it works for you I would still fine tune it depending on your configs, a lot of the time fine tuning samba is a pain and most of the time it actually comes down to crappy network performance I.E the nic seems to be working but it actually has a fault.
Regards Per Qvindesland
--- Original message follows --- Subject: Re: [CentOS] Samba "use sendfile" configuration option set do disabled as default - why? From: Christoph Maser cmr@financial.com To: "CentOS mailing list" centos@centos.org Date: 24-08-2009 13:27
Am Montag, den 24.08.2009, 13:24 +0200 schrieb Per Qvindesland:
socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536
These options are often found if you search for samba tuning. Did someone actaully benchmark the results? Shouldn't tuning buffers manuall be oboslete nowadays? I somewhat doubt this is THE solution for slow sambas.
Chris
financial.com AG
Munich head office/Hauptsitz München: Maria-Probst-Str. 19 | 80939 München | Germany Frankfurt branch office/Niederlassung Frankfurt: Messeturm | Friedrich-Ebert-Anlage 49 | 60327 Frankfurt | Germany Management board/Vorstand: Dr. Steffen Boehnert (CEO/Vorsitzender) | Dr. Alexis Eisenhofer | Dr. Yann Samson | Matthias Wiederwach Supervisory board/Aufsichtsrat: Dr. Dr. Ernst zur Linden (chairman/Vorsitzender) Register court/Handelsregister: Munich – HRB 128 972 | Sales tax ID number/St.Nr.: DE205 370 553 _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
On Mon, 2009-08-24 at 15:41 +0200, happymaster23 wrote:
Thank you for replies,
with option "use sendfile = yes" it works great (with good speed), but I am asking why is this option in CentOS disabled as default, even in standard samba build it should be (according to Samba release notes) enabled as default.
So my question is, why in CentOS´s Samba build is this option as default disabled. Is it experimental? Is it potentially dangerous? (Something like oplocks?)
--------------- Why is it disabled? You need to ask the "Samba Devel List" that. Just an example an i586 class machine with 4 Nics will saturate a gig E connection easily. Its all all about disk throughput and how many disc spindles you have. Maybe this will explain it to you! :-) Winblows related mostly.
Hint use "man smb.conf"
use sendfile (S) If this parameter is yes, and the sendfile() system call is supported by the underlying operating system, then some SMB read calls (mainly ReadAndX and ReadRaw) will use the more efficient sendfile system call for files that are exclusively oplocked. This may make more efficient use of the system CPU’s and cause Samba to be faster. Samba automatically turns this off for clients that use proto- col levels lower than NT LM 0.12 and when it detects a client is Windows 9x (using sendfile from Linux will cause these clients to fail).
Default: use sendfile = false
JohnStanley
This is not problem of disks, this is problem of Samba (operating systems in network are Windows XP and Windows 7 only, so this should not be problem caused by Windows).
I am asking because of risks and disadvantages. This is same as oplocks - it may be performance tweak, but it is potentially dangerous (data corruption).
Thank you
2009/8/24 JohnS jses27@gmail.com:
On Mon, 2009-08-24 at 15:41 +0200, happymaster23 wrote:
Thank you for replies,
with option "use sendfile = yes" it works great (with good speed), but I am asking why is this option in CentOS disabled as default, even in standard samba build it should be (according to Samba release notes) enabled as default.
So my question is, why in CentOS´s Samba build is this option as default disabled. Is it experimental? Is it potentially dangerous? (Something like oplocks?)
Why is it disabled? You need to ask the "Samba Devel List" that. Just an example an i586 class machine with 4 Nics will saturate a gig E connection easily. Its all all about disk throughput and how many disc spindles you have. Maybe this will explain it to you! :-) Winblows related mostly.
Hint use "man smb.conf"
use sendfile (S) If this parameter is yes, and the sendfile() system call is supported by the underlying operating system, then some SMB read calls (mainly ReadAndX and ReadRaw) will use the more efficient sendfile system call for files that are exclusively oplocked. This may make more efficient use of the system CPU’s and cause Samba to be faster. Samba automatically turns this off for clients that use proto- col levels lower than NT LM 0.12 and when it detects a client is Windows 9x (using sendfile from Linux will cause these clients to fail).
Default: use sendfile = false
JohnStanley
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
happymaster23 wrote:
This is not problem of disks, this is problem of Samba (operating systems in network are Windows XP and Windows 7 only, so this should not be problem caused by Windows).
I am asking because of risks and disadvantages. This is same as oplocks - it may be performance tweak, but it is potentially dangerous (data corruption).
How does sendfile() mesh with SMB signed packets? I'd think you would have to copy the data for that. Sendfile should just be saving a memory/buffer copy or two anyway. That seems like it should be a barely measurable amount of time unless your server is heavily loaded.
On Mon, 2009-08-24 at 19:01 +0200, happymaster23 wrote:
This is not problem of disks, this is problem of Samba (operating systems in network are Windows XP and Windows 7 only, so this should not be problem caused by Windows).
I am asking because of risks and disadvantages. This is same as oplocks - it may be performance tweak, but it is potentially dangerous (data corruption).
----- If you are using XP and Vista you should not need it. I did not say it was a disk problem either. Please bottom post all replies. All it really does is keep an in memory copy. If the server is heavily loaded you can start swapping witch will farther reduce performance. Also like the man page says read raw = yes can be set also with out sendfile. Problem, you really dont say what the exact problem is...Slow?
John
No, I have not any problem. With use sendfile = yes it works great. My question is, why is this configuration option set to be off in CentOS?
You know - when I want to read something about oplocks (advantages & disadvantages) it was very easy - there are tons of documentation. When I want to read something about use sendfile, it is big problem, becuse there is nothing.
My question is why is this option disabled as default in CentOS? And I am not alone with this question - http://fixunix.com/samba/184987-samba-red-hat-use-sendfile.html
It is two WD Caviar GP in SW RAID 1 and LVM2, so there can´t be a problem.
2009/8/25 JohnS jses27@gmail.com:
On Mon, 2009-08-24 at 19:01 +0200, happymaster23 wrote:
This is not problem of disks, this is problem of Samba (operating systems in network are Windows XP and Windows 7 only, so this should not be problem caused by Windows).
I am asking because of risks and disadvantages. This is same as oplocks - it may be performance tweak, but it is potentially dangerous (data corruption).
If you are using XP and Vista you should not need it. I did not say it was a disk problem either. Please bottom post all replies. All it really does is keep an in memory copy. If the server is heavily loaded you can start swapping witch will farther reduce performance. Also like the man page says read raw = yes can be set also with out sendfile. Problem, you really dont say what the exact problem is...Slow?
John
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
happymaster23 wrote:
My question is why is this option disabled as default in CentOS?
I believe you are not getting the answer you seek because you are not asking the right question to the right people.
The right question is "Why is this option disabled as default in *RHEL*?" The right people to ask this question is Red Hat.
I doubt the CentOS builders are arbitrarily changing this default setting, but feel free to grab the upstream srpm and show me that I'm wrong.
Rick
Rick Barnes wrote:
happymaster23 wrote:
My question is why is this option disabled as default in CentOS?
I believe you are not getting the answer you seek because you are not asking the right question to the right people.
The right question is "Why is this option disabled as default in *RHEL*?" The right people to ask this question is Red Hat.
I doubt the CentOS builders are arbitrarily changing this default setting, but feel free to grab the upstream srpm and show me that I'm wrong.
You are absolutely correct.
The reason every config file is set the way it is is because that is the way it is set in RHEL.
The whole point of CentOS is to produce a distribution that acts and feels like the applicable upstream version.
-
happymaster23 -
Johnny Hughes -
JohnS -
Les Mikesell -
Per Qvindesland -
Rick Barnes