About pppoe disconnected rate question

PPPoE related questions
jalee
Posts: 32
Joined: 01 Oct 2017, 16:42

About pppoe disconnected rate question

Post by jalee »

Dear All,
Can I control disconnected rate? as I know we can configure "connlimit", but how can I configure disconnected rate?
such as: please refer topology as below.
if we had 8K pppoe session abnormal link down on A_clients site, I will saw accel-ppp server cpu utilization will reach around 100% and the B_clients site will over 40% pppoe session disconnected, because these session LCP echo request timeout.

Can we possible control disconnected rate or priority of disconnect process?

the accel-ppp version is 1.11.2

Thanks,


topology:
(A_clients) 8K----X----(L2_switch)------accel-ppp
(B_clients) 8K--------------|


=============== accel-ppp.conf configure.
[connlimit]
limit=70/s
burst=3
timeout=30

[ppp]
lcp-echo-interval=90
lcp-echo-failure=2

[core]
log-error=/var/log/accel-ppp/core.log
thread-count=12
Dmitry
Администратор
Posts: 954
Joined: 09 Oct 2014, 10:06

Re: About pppoe disconnected rate question

Post by Dmitry »

no, there is no such feature
if you experience hangs during massive disconnect then you can try unit-cache
[ppp]
unit-cache=N - number of interfaces to keep in down state for later use
the fact is that kernel slowly destroy network interfaces and working threads gets blocked in waiting for interface destory

also consider to limit number of simultaneous requests to radius:
[radius]
server=x.x.x.x,req-limit=N
jalee
Posts: 32
Joined: 01 Oct 2017, 16:42

Re: About pppoe disconnected rate question

Post by jalee »

Got it.
let me try it.

Thanks!!
jalee
Posts: 32
Joined: 01 Oct 2017, 16:42

Re: About pppoe disconnected rate question

Post by jalee »

Dear Dmitry,
the unit-cache function can't help, the B_clients side still have many session LCP echo request timeout.

I found out other thing, if I normal hung up 8K session (it will take around 3~5 minute), on this moment I can't dial up any pppoe session.
we must be waiting 8K session disconnect succeeded then begin new pppoe session.
Please help check!

seemed the "hung up" process priority be hight to "create session" process.
can we control "hung up" and "create session" of process priority or weight?

Thanks,
Dmitry
Администратор
Posts: 954
Joined: 09 Oct 2014, 10:06

Re: About pppoe disconnected rate question

Post by Dmitry »

ok, i'll try to reproduce
Dmitry
Администратор
Posts: 954
Joined: 09 Oct 2014, 10:06

Re: About pppoe disconnected rate question

Post by Dmitry »

try to update from git master
there are 2 new commits that should improve responsiveness
so try following adjustments:
[ppp]
unit-cache=1 (or more)
lcp-echo-timeout=60
lcp-echo-interval=20
lcp-echo-failure=3

also make sure to set req-limit for radius server

i didn't try to reproduce massive disconnect tho as it is not easy...
jalee
Posts: 32
Joined: 01 Oct 2017, 16:42

Re: About pppoe disconnected rate question

Post by jalee »

Hi Dmitry,

I try to git clone accel-ppp and re-install again. the cli command is right?
"git clone git://github.com/xebd/accel-ppp.git"

I verify 1.11.2 and new git clone version, I found out if I add "lcp-echo-timeout" we can succeeded create some(not much) new pppoe session when we normal hung up 8K pppoe session.

but new git clone version, can't resolve abnormal link down on A_clients side issue and we also can't create new pppoe session, we must be waiting abnormal 8K session disconnect succeeded then begin new pppoe session.

can we possible fix abnormal link down situation?

please help check, I can try it if you have any idea.

Thanks.
Best Regards,
Dmitry
Администратор
Posts: 954
Joined: 09 Oct 2014, 10:06

Re: About pppoe disconnected rate question

Post by Dmitry »

this is main repository: git clone https://git.code.sf.net/p/accel-ppp/code
jalee
Posts: 32
Joined: 01 Oct 2017, 16:42

Re: About pppoe disconnected rate question

Post by jalee »

Hi Dmitry,
This version is almost same result, as before I said. the dial up rate about 4~5/ps.
I observe normal hung up 8K the cpu utilization around 96%, but abnormal link down is 100%.

give your referance.
in my lab:
test result record is,
condition (16K pppoe session, local AAA "pap", 4 cpu core, 4G ram, ubuntu16.04.3, 4.4.0-93-generic x86_64)
normal dial up rate is 26/ps
cpu utilization average rate is 85%.

====================== log
inet@250-50-client:~$ accel-cmd show stat
uptime: 0.01:40:24
cpu: 0%
mem(rss/virt): 166828/2162944 kB
core:
mempool_allocated: 20857319
mempool_available: 552071
thread_count: 4
thread_active: 2
context_count: 17004
context_sleeping: 0
context_pending: 0
md_handler_count: 32004
md_handler_pending: 0
timer_count: 15997
timer_pending: 0
sessions:
starting: 0
active: 16000
finishing: 0
pppoe:
starting: 0
active: 16000
delayed PADO: 0
recv PADI: 55076
drop PADI: 0
sent PADO: 55076
recv PADR(dup): 45245(4563)
sent PADS: 44653
filtered: 0
inet@250-50-client:~$

BR,
Dmitry
Администратор
Posts: 954
Joined: 09 Oct 2014, 10:06

Re: About pppoe disconnected rate question

Post by Dmitry »

do you use radius for authentication or chap-secrets ?
Post Reply