- CheckMates
- :
- Products
- :
- General Topics
- :
- Re: CoreXL: only one SND core is busy
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page
Are you a member of CheckMates?
×- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
CoreXL: only one SND core is busy
I'll put pictures because the copy paste breaks the table formatting:
fw ctl affinity -l -r
cpstat os -f multi_cpu (it's Sunday morning so not much traffic)
As you can see only the first SND (on the first CPU core) seems to be doing anything. The other SND (CPU 2/3 depending on the command) is mostly idling.
I suspect this may be because there is only one LACP bond interface (of two physical interfaces) on the gateway (but with many vlan subinterfaces). Therefore all the queuing of this bonding interface would maybe be handled by the first SND only, does that make sense?
If this is correct, I guess I should reconfigure CoreXL with only one SND?
It is an open server R77.30 gateway.
- Tags:
- corexl
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The SND/IRQ core is assigned to the physical interfaces, eth1 is being handled by CPU 0 while eth0 is being handled by CPU 2. 95% of your traffic is fully accelerated in the SXL path (and a relatively high templating rate of 49%) so higher CPU load is to be expected on your SND/IRQ cores, a few points:
- Is SMT/Hyperthreading enabled? (/sbin/cpuinfo) If so you need to disable it and then assign 4 kernel instances via cpconfig (which will assign 2 SND/IRQ cores). Having SMT enabled with such a high percentage of accelerated traffic actually hurts performance.
- Since it looks like all traffic is being handled by just two physical interfaces, I'd recommend enabling Multi-Queue on both of them even though the RX-DRP rate is well below 0.1%. This will help to spread the load out a bit among the 2 SND/IRQ cores.
- Your distribution of bonded traffic between the two physical interfaces looks pretty even, so no issues there.
- You definitely do NOT want to drop from 2 SND/IRQ cores down to one.
- I suppose you could configure 4 SND/IRQ cores w/ Multi-Queue enabled, not sure how much that would help since there are only two active physical NICs but it probably wouldn't hurt.
--
Second Edition of my "Max Power" Firewall Book
Now Available at http://www.maxpowerfirewalls.com
CET (Europe) Timezone Course Scheduled for July 1-2
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
You have 12 CPU cores, but in fact only 8 are licensed.
Anyway, have a look here if it helped:
Jozko Mrkvicka
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
You have 12 CPU cores, but in fact only 8 are licensed.
I know this but I don't see how it is related to the second SND idling.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Need full Super Seven command outputs to answer your question:
Super Seven Performance Assessment Commands (s7pac)
This will help answer questions such as how much traffic is accelerated, whether SecureXL is enabled, whether you are using manual affinity, etc.
--
Second Edition of my "Max Power" Firewall Book
Now Available at http://www.maxpowerfirewalls.com
CET (Europe) Timezone Course Scheduled for July 1-2
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Here are the missing 5:
fwaccel stats -s
Accelerated conns/Total conns : 7337/14761 (49%)
Accelerated pkts/Total pkts : 138590646/145791596 (95%)
F2Fed pkts/Total pkts : 5881723/145791596 (4%)
PXL pkts/Total pkts : 1319227/145791596 (0%)
QXL pkts/Total pkts : 0/145791596 (0%)
grep -c ^processor /proc/cpuinfo
12
netstat -ni (sorry it's quite long)
Kernel Interface table
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg
bond0 1500 0 1394995720953 5039 39663538 0 1379697498747 0 0 0 BMmRU
bond0.4 1500 0 52359489686 0 0 0 21412786937 0 0 0 BMmRU
bond0.5 1500 0 369278960681 0 0 0 747705040016 0 0 0 BMmRU
bond0.6 1500 0 62052221192 0 0 0 60992876777 0 0 0 BMmRU
bond0.7 1500 0 3713019794 0 0 0 18864528927 0 0 0 BMmRU
bond0.8 1500 0 0 0 0 0 1006311 0 0 0 BMmRU
bond0.10 1500 0 10606207916 0 0 0 7136356779 0 0 0 BMmRU
bond0.14 1500 0 26754476499 0 0 0 20827379092 0 0 0 BMmRU
bond0.19 1500 0 2382647063 0 0 0 706305695 0 0 0 BMmRU
bond0.24 1500 0 19065409562 0 0 0 14061884167 0 0 0 BMmRU
bond0.25 1500 0 294290424 0 0 0 302684345 0 0 0 BMmRU
bond0.26 1500 0 302497724 0 0 0 251089232 0 0 0 BMmRU
bond0.27 1500 0 116215093 0 0 0 110093772 0 0 0 BMmRU
bond0.29 1500 0 30457344 0 0 0 31915625 0 0 0 BMmRU
bond0.79 1500 0 891130915 0 0 0 1318803777 0 0 0 BMmRU
bond0.98 1500 0 3284038 0 0 0 651117 0 0 0 BMmRU
bond0.101 1500 0 119772512 0 0 0 137775320 0 0 0 BMmRU
bond0.107 1500 0 45142551 0 0 0 29757675 0 0 0 BMmRU
bond0.111 1500 0 8696900244 0 0 0 4742540248 0 0 0 BMmRU
bond0.116 1500 0 490016686 0 0 0 460636644 0 0 0 BMmRU
bond0.118 1500 0 4139898545 0 0 0 1977956603 0 0 0 BMmRU
bond0.121 1500 0 853925844 0 0 0 499164596 0 0 0 BMmRU
bond0.122 1500 0 179499941 0 0 0 132989328 0 0 0 BMmRU
bond0.170 1500 0 23396028007 0 0 0 28860775722 0 0 0 BMmRU
bond0.173 1500 0 8780230716 0 0 0 5201512913 0 0 0 BMmRU
bond0.211 1500 0 538112194 0 0 0 1180701623 0 0 0 BMmRU
bond0.216 1500 0 80064697232 0 0 0 44714847049 0 0 0 BMmRU
bond0.218 1500 0 718764 0 0 0 614903 0 0 0 BMmRU
bond0.249 1500 0 131502970 0 0 0 144575776 0 0 0 BMmRU
bond0.254 1500 0 84007379531 0 0 0 72725532974 0 0 0 BMmRU
bond0.255 1500 0 280525503 0 0 0 373980096 0 0 0 BMmRU
bond0.307 1500 0 0 0 0 0 612125 0 0 0 BMmRU
bond0.316 1500 0 338881243 0 0 0 354686563 0 0 0 BMmRU
bond0.349 1500 0 3261618818 0 0 0 3152789986 0 0 0 BMmRU
bond0.408 1500 0 6738301 0 0 0 6430679 0 0 0 BMmRU
bond0.409 1500 0 33830202905 0 0 0 15597198399 0 0 0 BMmRU
bond0.411 1500 0 45680844 0 0 0 27421501 0 0 0 BMmRU
bond0.413 1500 0 110364862743 0 0 0 53024348871 0 0 0 BMmRU
bond0.416 1500 0 5816040525 0 0 0 6348221543 0 0 0 BMmRU
bond0.417 1500 0 39405084138 0 0 0 15587313742 0 0 0 BMmRU
bond0.419 1500 0 3483246523 0 0 0 4422397258 0 0 0 BMmRU
bond0.507 1500 0 0 0 0 0 612685 0 0 0 BMmRU
bond0.511 1500 0 1580213 0 0 0 881229 0 0 0 BMmRU
bond0.518 1500 0 5093539 0 0 0 6175567 0 0 0 BMmRU
bond0.521 1500 0 414168783 0 0 0 650313999 0 0 0 BMmRU
bond0.611 1500 0 216808003 0 0 0 223187242 0 0 0 BMmRU
bond0.641 1500 0 199609888915 0 0 0 92413132506 0 0 0 BMmRU
bond0.643 1500 0 80757360 0 0 0 42352700 0 0 0 BMmRU
bond0.644 1500 0 761316 0 0 0 610334 0 0 0 BMmRU
bond0.645 1500 0 148104750 0 0 0 44926296 0 0 0 BMmRU
bond0.653 1500 0 10976675 0 0 0 7416242 0 0 0 BMmRU
bond0.661 1500 0 184047965 0 0 0 147412665 0 0 0 BMmRU
bond0.662 1500 0 0 0 0 0 611729 0 0 0 BMmRU
bond0.663 1500 0 0 0 0 0 616366 0 0 0 BMmRU
bond0.671 1500 0 99791137303 0 0 0 48503733230 0 0 0 BMmRU
bond0.672 1500 0 28107423 0 0 0 22814680 0 0 0 BMmRU
bond0.673 1500 0 19990806 0 0 0 23292324 0 0 0 BMmRU
bond0.674 1500 0 2819356727 0 0 0 1238707791 0 0 0 BMmRU
bond0.681 1500 0 26829948954 0 0 0 11069005509 0 0 0 BMmRU
bond0.682 1500 0 60479150 0 0 0 72024059 0 0 0 BMmRU
bond0.683 1500 0 224978500 0 0 0 306106657 0 0 0 BMmRU
bond0.691 1500 0 54724491309 0 0 0 26333668129 0 0 0 BMmRU
bond0.692 1500 0 1898751795 0 0 0 2639172349 0 0 0 BMmRU
bond0.694 1500 0 2728296755 0 0 0 1477418884 0 0 0 BMmRU
bond0.695 1500 0 869064 0 0 0 621226 0 0 0 BMmRU
bond0.696 1500 0 5275010566 0 0 0 2878140596 0 0 0 BMmRU
bond0.697 1500 0 4596758 0 0 0 4380694 0 0 0 BMmRU
bond0.698 1500 0 1951112 0 0 0 1826269 0 0 0 BMmRU
bond0.904 1500 0 13170157932 0 0 0 9857775878 0 0 0 BMmRU
bond0.905 1500 0 14251494963 0 0 0 10461430294 0 0 0 BMmRU
bond0.1025 1500 0 16777504 0 0 0 19842306 0 0 0 BMmRU
bond0.1026 1500 0 147591886 0 0 0 321596472 0 0 0 BMmRU
bond0.1032 1500 0 12563176 0 0 0 11953149 0 0 0 BMmRU
bond0.1080 1500 0 9763356910 0 0 0 14883147090 0 0 0 BMmRU
bond0.1099 1500 0 30982634 0 0 0 37281323 0 0 0 BMmRU
eth0 1500 0 740726327381 911 18724147 0 661416498362 0 0 0 BMsRU
eth1 1500 0 654269393572 4128 20939391 0 718281000385 0 0 0 BMsRU
lo 16436 0 44694202 0 0 0 44694202 0 0 0 LRU
fw ctl multik stat
ID | Active | CPU | Connections | Peak
----------------------------------------------
0 | Yes | 11 | 1765 | 18016
1 | Yes | 9 | 1240 | 17449
2 | Yes | 7 | 1346 | 29195
3 | Yes | 5 | 1182 | 29215
4 | Yes | 3 | 1735 | 14742
5 | Yes | 1 | 1585 | 14783
6 | Yes | 10 | 1740 | 57699
7 | Yes | 8 | 1633 | 16750
8 | Yes | 6 | 1088 | 19264
9 | Yes | 4 | 1477 | 18676
But the question is also, more generally speaking, in case of bonding is the SND assigned to the bonding interface or to the underlying physical interfaces?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The SND/IRQ core is assigned to the physical interfaces, eth1 is being handled by CPU 0 while eth0 is being handled by CPU 2. 95% of your traffic is fully accelerated in the SXL path (and a relatively high templating rate of 49%) so higher CPU load is to be expected on your SND/IRQ cores, a few points:
- Is SMT/Hyperthreading enabled? (/sbin/cpuinfo) If so you need to disable it and then assign 4 kernel instances via cpconfig (which will assign 2 SND/IRQ cores). Having SMT enabled with such a high percentage of accelerated traffic actually hurts performance.
- Since it looks like all traffic is being handled by just two physical interfaces, I'd recommend enabling Multi-Queue on both of them even though the RX-DRP rate is well below 0.1%. This will help to spread the load out a bit among the 2 SND/IRQ cores.
- Your distribution of bonded traffic between the two physical interfaces looks pretty even, so no issues there.
- You definitely do NOT want to drop from 2 SND/IRQ cores down to one.
- I suppose you could configure 4 SND/IRQ cores w/ Multi-Queue enabled, not sure how much that would help since there are only two active physical NICs but it probably wouldn't hurt.
--
Second Edition of my "Max Power" Firewall Book
Now Available at http://www.maxpowerfirewalls.com
CET (Europe) Timezone Course Scheduled for July 1-2
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thank you very much that answers all my questions
