sm00thpapa
Very Senior Member
I saw this last night with .1354. I went to .380.1681 and it went away.
[Bug] CPU loading is 100% in core 1
- Thread starter vanic
- Start date
Hi all,
Thanks for your help, we have found some issues about ISP Movistar triple vlan setting.
1. CPU loading issue
2. VPN issue
3. qos vs. IPTV issue
4. others
We have fixed some bugs and will check with Movistar about some setting for users. Thanks you guys a lot!
Thanks,
Vanic
Thanks for your help, we have found some issues about ISP Movistar triple vlan setting.
1. CPU loading issue
2. VPN issue
3. qos vs. IPTV issue
4. others
We have fixed some bugs and will check with Movistar about some setting for users. Thanks you guys a lot!
Thanks,
Vanic
Which is not surprising at all, although I haven't looked at much code that uses this subsystem so what I say probably isn't accurate.
The soft irqs are probably from network IO and the mechanism most commonly used by drivers is to service the irq and delegate the work to a soft irq handler. These handlers are guaranteed to run on the same cpu that queued them to prevent ordering problems and simplify locking. So you should expect to see a single CPU getting hit.
That's by kernel design of the soft irq handler mechanism.
I don't know about which CPU irqs are delivered, maybe only one of the CPUs, but that's possibly hardware design specific, don't know.
So there might not be much that can be done about that.
Looking at the posts above I suspect the question to be answered is why is the softirq handler for cpu 0 not working as hard when QoS enabled given the reported cpu usgae (the sirq) is still relatively high.
Perhaps one of the drivers is blocking during servicing the IO for some reason causing throughput drop.
I'd be checking what changes have been made that could affect the driver(s) soft irq handler performing the IO if it was my problem, in particular changes to locking of locks used by both driver(s) and other areas of the subsystem.
It probably isn't viable with a router but oprofile should be able to identify where the time is being spent which could give you a fairly specific area of code to survey.
Ian
From my observation on the broadcom contaminated kernel, I think you're correct. The issue of this thread is perhaps less interesting. Might be simply some user space apps hogging on core 0...
You brought up one interesting point. I wonder for IPSec.. can the encryption/encapsulation be done asynchronously on core 1 while packet flow process still handled by core 0?
In the current broadcom contaminated 2.6 kernel, it's all done in core 0, sadly.
Yeah, but that design is vanilla kernel, driver writers use it as it's the recommended way to do it.
You'd think so, but perhaps it's done as part of the packet receive, don't know how that stuff fits together.
one question, because speak with askey, askey is Asus computer. They do the software and manufacturin the new router of Movistar, they do the software specific of Movistar, and resolved moré problems. I have the latest firmware mitrastar and comtrend is two of the routers that puts Movistar today, are you interested?
Last edited:
Hi rubencc33,
They do the software and manufacturin the new router of Movistar. => Really? I don't know about that, this case isn't my task, I shouldn't interrupt even if I am interested.
Thanks,
Vanic
as you said you have to talk to Movistar, topic of what profiles, but we wondered if you wanted the firmware than routers Movistar, if you served for help with something, and if askey is the new manufacturer delos Home Gateway unit Movistar
Hi Rubencc33,
Thanks for your kind suggestion, if any issue, we will contact with Movistar directly by ASUS (won't via Askey). Actually, we always keep contacting with Movistar (the contact window isn't me.), you don't worry about anything. As I said, this case is my colleague's, I don't want to interrupt, I think my colleague will handle that.
Thanks,
Vanic
I imagine you'll have to talk to Movistar, I mean that I have in my possession the firmware two routers Movistar being installed was only if you wanted or your partner so they were sent around, my english is a bit bad and I have not been able to explain, sorry and thank you very much
trujulu
Occasional Visitor
Yes, I am having it another time. Core 2 is not being used.
I have this issue with any 380 fw. And the download speed is maximum 850 Mbps on a gigabit connection. With latest 378 from asus I don't have this problem (with only one cpu being used) and the speed is normal, around 930 Mbps.
Ps : using an ac56u router.
Sent from my SM-G930F using Tapatalk
Ps : using an ac56u router.
Sent from my SM-G930F using Tapatalk
Sorry to dig this one up, I am on latest firmware of ASUS RT AC68U (380.7743) and this is happening to me.
Core 1 = 99%
Core 2 = 2-6% (at the most)
I am on a gigabit ethernet connection from my ISP (MEO-Portugal) 1000mbps/200mbps and can only hit like 850/200.
Core 1 goes bananas and core 2 is asleep.
Still not fixed?
PS: I have Revision E1 of the Router, so its 1ghz cpu
Core 1 = 99%
Core 2 = 2-6% (at the most)
I am on a gigabit ethernet connection from my ISP (MEO-Portugal) 1000mbps/200mbps and can only hit like 850/200.
Core 1 goes bananas and core 2 is asleep.
Still not fixed?
PS: I have Revision E1 of the Router, so its 1ghz cpu
Sorry can you explain a little better? "network stack" ?
I was told this router could reach 940mbps but its hitting 99% core usage on core 1
ColinTaylor
Part of the Furniture
He's referring to the networking code that is part of the router's Linux kernel. Any traffic that is entirely processed by this code will be limited by the power of the router's CPU. To achieve maximum throughput the router needs to have "hardware acceleration" enabled which bypasses the normal network stack and uses proprietary features of the Broadcom chipset.
And how can I enable hardware acceleration if you dont mind me asking?
