Pentium 4D 2,8GHz, 512 Ram dual e1000 net adapter. My problem is high load - when i have about 500 sessions active i get about 90% of cpu usage. On pppoe interfaces i have simple queues added. All works with from remote radius aaa. Generally my question is about load - is this level of cpu usage is normal?
Does the performance of the network degrade? if not then no worries, you could add a 2nd server to the same ethernet collision domain then PPPoE sessions would be distributed between them.
Earlier I had 3com eth cards, load was 100% - bad things happen ;> I’m thinking about second machine, but I think that this one shoud easily handle 1000 sessions 
So I’m asking if someone has similar configuration to compare loads.
So what type of cards are you using now? Intel server ones are best I’ve found so far and are what others have also reported as good, I have had good success with Compaq (intel chipset) NICs bought from eBay
Intel dual port server adapter pci-e (e1000).
newtonek, set Change TCP MSS to no to reduce CPU utilization
I don’t understand ;p
Change TCP MSS is set to yes
Pentium 4D 2,8GHz, 512 Ram dual e1000 net adapter. My problem is high load - when i have about 500 sessions active i get about 90% of cpu usage. On pppoe interfaces i have simple queues added. All works with from remote radius aaa. Generally my question is about load - is this level of cpu usage is normal?
Under “Queue”, “Queue Types” what is your default queue type? If its not pfifo I would change it to pfifo.
We use Intel PRO/100 network cards integrated on Supermicro P4 motherboards. Around 500 PPPoE sessions per server and no noticeable CPU load.
Matt
“Change TCP MSS” is a setting in your ppp profile
I’ve had pfifo, now it’t bfifo for test. Same load. When ypu have 500 sessions what’s your load, what machine? You have simple queues on ppp interfaces?
Its not peak time now. 337 active connections doing a total about 5mbps between them all. Most are set at 512kbps. PPPoE automatically adds the queues.
system resource> print
uptime: 3w5d1h28m49s
version: “2.9.39”
free-memory: 335892kB
total-memory: 387076kB
cpu: “Intel(R)”
cpu-frequency: 2499MHz
cpu-load: 9
free-hdd-space: 10307012kB
total-hdd-space: 10365292kB
write-sect-since-reboot: 48490720
write-sect-total: 372666360
Matt
I have about tx 25 rx 12 Mbps. Maby this is the case.
Hmm, could be. Use Mikrotik firewall to drop all p2p and see if the problem goes away. ;<) You might try a couple of good quality Intel NIC’s. Likely only 30$ a piece so cheap to try.
http://www.newegg.com/Product/Product.asp?Item=N82E16833106123
Before ordering anything make sure Mikrotik supports the model.
Matt
Users pay to suck ;] I have very good NIC - Intel dual port gigabit server adapter. Dropping p2p is not possible
in your PPP profiles, set Change TCP MSS to no for all profiles assigned to your users (secrets), this will reduce CPU utilization to a significant level
Set your TCP MSS from ip > firewall > mangle.
choose forward chain
protocol = tcp(6)
tcp flags = syn
action = change MSS
new TCP MSS = 1400 Rafiq…
http://forum.linux.org.bd
Hmm. It may really work, now i have about 30% load. I’ll see in the evening.