More than 80% cpu usage = lag spikes
Everyone thanks a lot for your feedback regarding this issue.
I went into the datacenter today and had another good look at the BIOS
settings. In the end I did a full BIOS reset and re-programmed all the options.
Servers now run at a rock solid 250FPS at around 80-90% CPU load per 24max
server per core (acceptable).
Saint K.
________________________________________
From: hlds_linux-boun...@list.valvesoftware.com
[hlds_linux-boun...@list.valvesoftware.com] On Behalf Of Svensk Ljud& Ljus
Produktion [i...@teaterljud.se]
Sent: 25 July 2011 17:00
To: hlds_linux@list.valvesoftware.com
Subject: Re: [hlds_linux] cpu on i7 920
To get a true picture on the cpu load you have to monitor both the
machines cpu-usage by core, and the srcds instances usage of cpu.
This can be monitored by munin.
Offcource it can also monitor fps, no.off players, uptime and the
network traffic on the nic by port.
But dont forget to monitor the memory usage on the machine.
Its a easy way to find errors.
Peter
Sweden
Eric Riemers skrev 2011-07-25 15:04:
I only have issues on the box with the 32 slots server, and since tf2 is
build for 24 i dont think there will be any performance increases in this
area. The 24 slots sit around 60/70% and i've got no complaints from there
(except for the few that always complain)
Also if you have hlxce running or similar you can usually see your
serverside fps too, not that this matters that much but if you see drops
below 66 it could be cpu, at least a reference point to look at if people
complain about lag.
Still running an older distro of debian on that box though, perhaps a
upgrade would do some good i presume too.
On Mon, 25 Jul 2011 14:05:10 +0200, Saint K.<sai...@specialattack.net>
wrote:
Hi,
My concern is not really to have more servers on there (although it be
nice
if possible), but I'd like to get at least 66+fps stable per server per
core, and having some load left to have replay etc enabled, as we had to
kill that as well to get things running on the edge of normal.
I've also tried assigning it per core, but never found clear bennifits to
do so. The keeping one core free is just the thought process, leaves some
room for other processes to peak (if required).
Saint K.
________________________________________
From: hlds_linux-boun...@list.valvesoftware.com
[hlds_linux-boun...@list.valvesoftware.com] On Behalf Of
gamead...@127001.org [gamead...@127001.org]
Sent: 25 July 2011 13:56
To: 'Half-Life dedicated Linux server mailing list'
Subject: Re: [hlds_linux] cpu on i7 920
Regardless of CPU usage you're going to struggle running more servers
than
that.
The thing is, it's not that the servers use a lot of CPU*, it's just that
when they need it, they need it _NOW_, or else they miss their "window"
and
the framerate drops. The more things running per core, the higher the
chance that something else will stop a server getting its slice when it
needs it.
*actually, they _DO_, but the point would stand even if they didn't
This is why I find load average useful; it can point out overloading (if
LA
number of threads your server has) even when the CPU load isn't showing
100%
Re: assigning servers to cores - this is going back about 3 years, but I
noticed that gameservers liked to jump cores about every 30 seconds or
so...
but if I pinned them to a core, they'd lag for a split-second around
every
30 seconds or so instead. No idea if that was the OS or the server, and
we've not tested again since, but we've had little issue leaving them to
roam free.
-----Original Message-----
From: hlds_linux-boun...@list.valvesoftware.com [mailto:hlds_linux-
boun...@list.valvesoftware.com] On Behalf Of Saint K.
Sent: 25 July 2011 12:22
To: Half-Life dedicated Linux server mailing list
Subject: Re: [hlds_linux] cpu on i7 920
I could live with that, at least knowing a reason why my 3000,- euro
hardware can only sustain 7-ish 24 slots TF2 servers.
Saint K.
________________________________________
From: hlds_linux-boun...@list.valvesoftware.com [hlds_linux-
boun...@list.valvesoftware.com] On Behalf Of James Botting
[bottswan...@googlemail.com]
Sent: 25 July 2011 13:18
To: Half-Life dedicated Linux server mailing list
Subject: Re: [hlds_linux] cpu on i7 920
Everytime a new weapon goes 'pew pew' the laser destroys a section of
your
CPU.
It's a new feature.
On 25/07/2011 12:14, "Andres Pozos"<javato...@yahoo.es> wrote:
Its not a bug its a "feature", jk. Orangebox take teh 100% cpu usage
in
a core with more than 25 slots, i tested it on many linux distros,
many
kernel configurations and many cpus. So welcome to the club.
Hi,
Thanks for the reply.
The servers are build according to Tyans best practise, so
everything
is inserted in the correct slots. I've recently also updated the BIOS
versions to be sure.
One thing I notice, at the memory settings is a "snooping" option -
If
this is disabled, the overall load is slightly lower (as it is now).
I'd generally wouldn't blame the hardware either, however, seeing I
can't find anything software wise, it's the logical next thing to
look
at.
Any more tips are welcome!
Saint K.
________________________________________
From: Jesse Molina [je...@opendreams.net]
Sent: 25 July 2011 12:36
To: Half-Life dedicated Linux server mailing list
Cc: Saint K.
Subject: Re: [hlds_linux] cpu on i7 920
Looks fine, don't mess with it.
generic-receive-offload would be good if you were doing 10G
networking,
but otherwise forget about it.
Make sure that your RAM is in the right slots as recommended by
Tyan.
That can slow things down sometimes but I would not expect that to
cause
such significant problems.
I have no clue. Grab an old cruddy desktop, set it up on your home
network, do a quickplay qualified server, and then watch how it
runs.
Use the same OS and versions you are using on your server and then
start
twiddling. You only need about a 2Mbps upstream rate for a 24-
player
server.
I would blame software way before I started blaming hardware.
Actually, I'd blame something you did unknowingly, first, but just
because I'm a BOFH.
People like to throw switches in the desperate hope that one of them
was
put there by system developers specifically just to slow things
down,
like a turbo switch on an old 486DX.
"Surely, my sheer desire to make things go faster and by randomly
throwing every bios setting, recompiling my kernel with obscure
realtime
patches I found, enabling weird sysctrl parameters, and buying a
Bigfoot
gaming NIC will make things go faster!"
And that's why we have fps_max 1000 and 100PPS update/cmd rates on
servers.
I really have no idea what I'm talking about. I've only been
messing
with srcds servers for the last nine months or so.
Then again... Seagate did ship me all of those SATA 300 drives with
the
150-limiting jumpers on by default.
Saint K. wrote:
Hi,
I am getting these values returned, not entirely sure what I am
looking at;
mrblonde:~# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: off
udp-fragmentation-offload: off
generic-segmentation-offload: on
generic-receive-offload: off
large-receive-offload: off
ntuple-filters: off
receive-hashing: off
Does this appear to be good? I've checked the chipset specs and it
supports checksum offloading.
Saint K.
________________________________________
From: hlds_linux-boun...@list.valvesoftware.com
[hlds_linux-boun...@list.valvesoftware.com] On Behalf Of Andrew
Armitage [and...@thirdlife.org]
Sent: 25 July 2011 10:36
To: hlds_linux@list.valvesoftware.com
Subject: Re: [hlds_linux] cpu on i7 920
I wouldn't expect problems with. But I happily admit to being no
expert.
Try ethtool -k<interface> and see what's what?
A
On 25/07/2011 08:42, Saint K. wrote:
The servers are build on Tyan Tempest i5400 motherboards, based on
the Intel 5400B chipset platform, the Gbit nic's used on this
board
are Intel 82563EB chips.
I've never really figured the load could be related to the
networking
chip as our throughput tests never really show any issues when
tested
(with all sorts of packet sizes, tcp/udp)
Saint K. ________________________________________ From:
hlds_linux-boun...@list.valvesoftware.com
[hlds_linux-boun...@list.valvesoftware.com] On Behalf Of Andrew
Armitage [and...@thirdlife.org] Sent: 25 July 2011 09:36 To:
hlds_linux@list.valvesoftware.com Subject: Re: [hlds_linux] cpu on
i7
920
Maybe the issue is the network hardware? HLDS is very network
intensive.
I believe that some cards support checksum offloading and some
don't.
A
On 25/07/2011 07:28, Saint K. wrote:
What I am still not getting is that our Xeon E5420's are doing
like 70-80% load on a single core for 24 players, and our Xeon
E5410's 90%+, where you say your older 4600+ does 70%.
Tried all sorts of different kernels out there.
Is there perhaps certain BIOS settings which could benefit when
running gameservers on them?
Ours surely should perform much better then that?
Saint K. ________________________________________ From:
hlds_linux-boun...@list.valvesoftware.com
[hlds_linux-boun...@list.valvesoftware.com] On Behalf Of Jesse
Molina [je...@opendreams.net] Sent: 25 July 2011 05:15 To:
Half-Life dedicated Linux server mailing list Subject: Re:
[hlds_linux] cpu on i7 920
I have a AMD Phenom x6 1055T doing multiple servers at the same
time. TF2 causes the active core to go to 100% for about two
seconds during map changes, but otherwise I've never seen it go
that high for extended periods of time. Average during full
24-player usage is about 40%.
I also have an older AMD Athlon64 x2 4600+ that runs a single
24-player TF2 quickplay server. It averages 70% usage when full
and gameplay is great.
Both of these are desktop class boards with DDR2 PC800M RAM.
Linux 2.6.39.
Try watching your CPU usage at (relatively) high resolution with
something like "htop -d 1"
No clue why you are maxing out like that.
Eric Riemers wrote:
All,
I run a 32 slots server on a i7 920 @ 2.67ghz, but i can see
with top and such that its potentially maxing out at 100% at
times since it only uses one core. Is it really now doing so
much
cpu that even a i7 core isn't enough? Didn't have much
complaints
before the pew pew.
24 slots on the same server do around 60% when full.
Eric
_______________________________________________ To unsubscribe,
edit your list preferences, or view the list archives, please
visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
-- # Jesse Molina # Mail = je...@opendreams.net # Page =
page-je...@opendreams.net # Cell = 1.602.323.7608 # Web =
http://www.opendreams.net/jesse/
_______________________________________________ To unsubscribe,
edit your list preferences, or view the list archives, please
visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________ To unsubscribe,
edit your list preferences, or view the list archives, please
visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________ To unsubscribe,
edit
your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________ To unsubscribe,
edit
your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list
archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list
archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
--
# Jesse Molina
# Mail = je...@opendreams.net
# Page = page-je...@opendreams.net
# Cell = 1.602.323.7608
# Web = http://www.opendreams.net/jesse/
_______________________________________________
To unsubscribe, edit your list preferences, or view the list
archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
