HP ProLiant DL360 Gen 10 Bios 4 cores with or without HT on 16 core system (8 cores with out HT)?

[Daniel_Kavan]

It looks like the default is General Power Efficient compute.   Does any one use General Peak Frequency or General throughput compute?  I'm considering changing to General throughput compute for Gaia R81.20.  Also, I don't see an option in the bios to turn Intel (R) Turbo Boost Technology off (it's grayed out).  Even after I switch to General Throughput compute, it's still grayed out.

[PhoneBoy]

Not sure what of those BIOS profile we test with.
I believe "Turbo Boost" applies to Hyperthreading...are you wanting to disable that?
You can disable it in cpconfig (at least on Check Point appliances) per sk93000.

[Daniel_Kavan]

It's hard to tell if hyperthreading is recommended for open servers or not, but it looks like it is supported after R80.40.

I'll test General throughput compute for Gaia R81.20 and get back to the community on how it goes.

[PhoneBoy]

In the past, there was an issue with licensing on Open Server when HT was enabled (namely we were treating HT cores as real cores from a licensing point of view).
That might have led to the decision to disable it on Open Servers (at least when R80.40 was released).
Believe that issue has been addressed: https://supportcenter.checkpoint.com/supportcenter/portal?eventSubmit_doGoviewsolutiondetails=&solut...

Given it's the default on most Check Point appliances now, it's probably safe to have it enabled in most cases on Open Server as well (with the caveats in sk93000 in mind).

[Daniel_Kavan]

I wonder which would have better performance optimization running Gaia R81.20 on an open server with 8 cores (16 with HT turned on).   8 cores with HT (8 are unused/other due to CP licensing) or 4 cores without HT (4 unused/other due to CP licensing)?

[_Val_]

SMT / HT will give you some performance advantage if all logical cores are licensed. Mind you will need to double the licensed cores in this case. How many cores are allowed by your license?

[Magnus-Holmberg]

U will run in to some issues.
This is a 16 Core machine with 8 Core licens.

So as you see here CPU 1-7 is assigned to multiQ, so at first look it looks awesome.
Like free performance... Yee!

And then you check like the VS allocation.
And here you see the crap then, meaning u only have 1 core for the firewalls.



Fixed with changing the fwk

fw ctl affinity -s -d -fwkall 6          and reboot.


You are not allowed to add 2 x 8 core licenses.
But oyu can ofc change it to a 16 core later if more umf is needed.

Regards
Magnus

[Daniel_Kavan]

Thanks, this looks like multi-queue isn't enabled on some interfaces.  How do I do that?  I found the command to set an interface to use MQ, but I'm getting an error.   Some drivers aren't supported, maybe that's it?  Update: yeah it doesn't look like tg3 is supported RE: MQ.  Well, that explains it, that's gtk.

fw ctl multik dynamic_dispatching get_mode
Current mode is on

[Expert@gw1:0]# fw ctl affinity -l -a
Interface eth4: CPU 2
Interface eth6: CPU 3
Interface eth7: CPU 4
Interface eth8: CPU 5
Interface eth10: CPU 1
Kernel fw_0: CPU 7
Kernel fw_1: CPU 6
Kernel fw_2: CPU 5
Kernel fw_3: CPU 4
Kernel fw_4: CPU 3
Kernel fw_5: CPU 2
Daemon cprid: CPU 2 3 4 5 6 7
Daemon mpdaemon: CPU 2 3 4 5 6 7
Daemon fwd: CPU 2 3 4 5 6 7
Daemon rad: CPU 2 3 4 5 6 7
Daemon lpd: CPU 2 3 4 5 6 7
Daemon core_uploader: CPU 2 3 4 5 6 7
Daemon wsdnsd: CPU 2 3 4 5 6 7
Daemon usrchkd: CPU 2 3 4 5 6 7
Daemon in.asessiond: CPU 2 3 4 5 6 7
Daemon pepd: CPU 2 3 4 5 6 7
Daemon pdpd: CPU 2 3 4 5 6 7
Daemon in.acapd: CPU 2 3 4 5 6 7
Daemon cp_file_convertd: CPU 2 3 4 5 6 7
Daemon iked: CPU 2 3 4 5 6 7
Daemon vpnd: CPU 2 3 4 5 6 7
Daemon cprid: CPU 2 3 4 5 6 7
Daemon cpd: CPU 2 3 4 5 6 7
Interface eth0: has multi queue enabled
Interface eth1: has multi queue enabled
Interface eth2: has multi queue enabled
Interface eth3: has multi queue enabled

[Expert@gw1:0]# mq_mng -o
Total 8 cores. Available for MQ 2 cores
i/f driver driver mode state mode (queues) cores
actual/avail
------------------------------------------------------------------------------------------------
eth0 i40e Kernel Up Auto (2/2) 0,1
eth1 i40e Kernel Up Auto (2/2) 0,1
eth2 i40e Kernel Up Auto (2/2) 0,1
eth3 i40e Kernel Up Auto (2/2) 0,1

[Expert@gw1# fw ctl affinity -l -r
CPU 0:
CPU 1: eth10
CPU 2: eth4
fw_5
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
CPU 3: eth6
fw_4
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
CPU 4: eth7
fw_3
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
CPU 5: eth8
fw_2
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
CPU 6: fw_1
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
CPU 7: fw_0
cprid mpdaemon fwd rad lpd core_uploader wsdnsd usrchkd in.asessiond pepd pdpd in.acapd cp_file_convertd iked vpnd cprid cpd
All:
Interface eth0: has multi queue enabled
Interface eth1: has multi queue enabled
Interface eth2: has multi queue enabled
Interface eth3: has multi queue enabled

[Daniel_Kavan]

In summary,

 I don't think this will be effective if those interfaces  (tg3 driver) are NOT set for MQ 
fw ctl affinity -s -d -fwkall 6   and reboot.  Also, is there a way to revert that command if it doesn't work?

See attached.  Do I have better performance with HT and 16, (12 cores in other) or is there a more optimal configuration?

Also, my understanding now is that the affinity of each interface should be set to one of the SND CPUs.   RE: using sim affinity -s In the case of 4 cores, you don't have to worry about it, since there is only one SND.  The SND will then allocate to one of the FW cores.

On an 8 core (16 with HT) system and with a license of 4 core, enabling HT doesn't provide any benefit only confusion.   4 cores are used, see attached.    My theory is that if I don't use HT, the 4 cores that are used will be faster than if I split them in 2 with HT.