RouterOS 3.0 final??? WOOT!

yes it is :>

when add 802.11n driver ?
eg AR5416 、AR5418

The version number doesn't concern me too much, its been working good. But I do have a question...

Did the multiple CPU and the L7 filter problems get resolved or are we still advised to turn those features off? These are problems documented and responded to my MT staff in the last 13/14 versions.

Thanks,
Scott

The version number doesn't concern me too much, its been working good. But I do have a question...

Did the multiple CPU and the L7 filter problems get resolved or are we still advised to turn those features off? These are problems documented and responded to my MT staff in the last 13/14 versions.

Thanks,
Scott

+1
Official release with officially unstable multicore support - it's a bad joke

The version number doesn't concern me too much, its been working good. But I do have a question...

Did the multiple CPU and the L7 filter problems get resolved or are we still advised to turn those features off? These are problems documented and responded to my MT staff in the last 13/14 versions.

Thanks,
Scott

+1
Official release with officially unstable multicore support - it's a bad joke

+1 multi-cpu

At least it's final.
Will be some time till we'll get it all smooth and running in every possible situation.
For me it works in just about every case.
Anyway, rc14 had the most significant amount of changes since rc1.
If you say it's "good to go", our trust is rising ....

The version number doesn't concern me too much, its been working good. But I do have a question...

Did the multiple CPU and the L7 filter problems get resolved or are we still advised to turn those features off? These are problems documented and responded to my MT staff in the last 13/14 versions.

Thanks,
Scott

+1
Official release with officially unstable multicore support - it's a bad joke

+1 multi-cpu

+1 - definitely interested in this too

What multi-core have to do with Mikrotik??????They only use Linux Kernel - not write it.

For those who are not aware of situation lately there are problems with latest Linux Kernel and SMP (multi-core). Some of the hardware is working fine some not. AFAIK problems are with drivers

I tried standard Linux web-proxy on the same box as my Mikrotik web-proxy and results was the same - it crashes.

only solution at this point is to find Linux open source drivers where everything is fine with SMP, but in Mikrotik with the same Kernel and same hardware there are problems - send to Mikrotik those drivers.

So far I am unable to locate such drivers for that particular box, so i use other (AMD,ATI)- no problems.

What multi-core have to do with Mikrotik??????They only use Linux Kernel - not write it.

For those who are not aware of situation lately there are problems with latest Linux Kernel and SMP (multi-core). Some of the hardware is working fine some not. AFAIK problems are with drivers

I tried standard Linux web-proxy on the same box as my Mikrotik web-proxy and results was the same - it crashes.

only solution at this point is to find Linux open source drivers where everything is fine with SMP, but in Mikrotik with the same Kernel and same hardware there are problems - send to Mikrotik those drivers.

So far I am unable to locate such drivers for that particular box, so i use other (AMD,ATI)- no problems.

You are right.

But. We are using RouterOS like a complete solution, but not as a driver.
We got the prepared package and want to utillize this package but not collect it.

You have complete package with multi-cpu=no by default.

Think about multi-cpu as additional feature, witch may or may not work because of limitations of Linux Kernel.

... or you prefer that they will take this "incomplete feature" out for good?


Idea about finding drivers were a little bit different, but I guess there is still popular illusion that for 35$ you should get everything without even a drop of sweat, so I guess there is no point to argue.

OSPF locked up on a RB333 2 hours after installing, requireing a reboot to before routes came back online, however I forgot to make a supout...

ForePoint,
it would be great, you
can generate support output file before reboot, if there will be the same problem with OSPF.
Send it to support@mikrotik.com
Thank you !

Idea about finding drivers were a little bit different, but I guess there is still popular illusion that for 35$ you should get everything without even a drop of sweat, so I guess there is no point to argue.

Hey, take it easy!

This is a technical Forum.
I don't discuss the price of the RouterOS.
I like this System. I'm ready to pay more for this System.
But in this case we wanna to solve problem with an additional feature of this Great System.

Is it Ok?

Forepoint,

Was your OSPF issue related to v3 also present in RC14? Has anyone else used OSPF in v3 and not had a problem. I want to upgrade my routers but OSPF operation is required. I am using RC14 without issues currently. There was nothing in the change log for ospf from RC14 to v3 so there should not have been any changes to cause problems not already in RC14.

Scott

Hey Normis... Thanks for the post. Personally, the multicore issue I don't care about as my cpu is less than 10% all the time anyway with about 50MB of traffic through it.

I am interested in hearing about the L7 status though and what your thoughts are on having it solid / or if you believe it works fine currently. There have been some posts in which MT staff has recommended to remove any L7 rules in order to fix the indicated problem. I understand this is a new feature and is also not required - we appreciate MT being as aggressive as they are to implement new features based on user requests, you all have done a superb job. We (I) just like to know what to use freely without issue and what I might do to cause myself a problem which is the only reason I asked for clarification on these two items I knew about.

Thanks Normis,

Scott

you have to understand that RouterOS uses the linux kernel which currently has this multicore issue. We are giving you RouterOS with singlecore mode by default, and you can try this multicore if you wish. We are just giving you this option to try it. We could ignore this feature altogether of course.

I understand you. Thank you for your great job.
I'm really like the RouterOS.
I will just waiting for RouterOS with the stable SMP module of linux kernel.

It's very bad when you pay for an L6 license and can't use ROS for full, because one CPU is not enough...
multi-cpu should be fixed.

There are big problems with latency to clients with ccq less than 60 %. The are big packet lost on v 3.0. On 2.9.x It was No problem. And there are problem in compacibility with older AP like the Ovislink AP 1120. The client is connect but it comunicate just the few seconds.

there is not one reasony why one core would not be enough. either your configuration is flawed, or you just haven't tested the actual performance of a good cpu.
even the rb1000 can handle any setup you can imagine!

I have an existing servers with intel dual/quad core CPUs for OSPF and PPPoE access with shaping and port filtering.
On one core it shows 70% load with 300 users.
Things in use:
- OSPF
- bridge filtering for rate-limiting PADIs
- PPPoE servers
- radius
Which conf can be flawed here?

there is not one reasony why one core would not be enough. either your configuration is flawed, or you just haven't tested the actual performance of a good cpu.

even the rb1000 can handle any setup you can imagine!

with any bandwith and any number of queue rules?

what cpu usage should I have on prescott 3.0 with 100 rules in mangle, 100 rules in filter, 10 rules in queues and 250Mbit/25k packets/s in both directions?

what cpu usage should I have on ...

does it matter? the real question here is - can it handle the tasks and not slow down the traffic. it can.

well, i has described real router which have cpu usage in such conditions more than 90% and it slow down traffic

RouterOS 3.0 final??? WOOT!

I just want to say 3.0 is not ready. more problem in 3.0

well, i has described real router which have cpu usage in such conditions more than 90% and it slow down traffic

I Disagree - while it is not 100% for more than a moment (not a spike) , everything is working on max.

IMHO CPU load option should displayed as "load" instead of percents. CPU load is a background value so there are no point for it to have 100 different positions, we need just 5 for example.

Something like:
0% - 15% - Idle load
16% - 40% - Low load
40% - 80% - Normal load
81% - 95% - High Load
96% - 100% - Full Load

And only thing you can see is those names "Idle", "Low" , "Normal", "High" and "Full".

I just want to say 3.0 is not ready. more problem in 3.0

I disagree, It is excellent "piece of software". MT Stuff just need to fix Packet Loss which occuring when having 160+ pppoe tunnels.

There are big problems with latency to clients with ccq less than 60 %. The are big packet lost on v 3.0. On 2.9.x It was No problem. And there are problem in compacibility with older AP like the Ovislink AP 1120. The client is connect but it comunicate just the few seconds.




I Have the same Problems

There are big problems with latency to clients with ccq less than 60 %. The are big packet lost on v 3.0. On 2.9.x It was No problem. And there are problem in compacibility with older AP like the Ovislink AP 1120. The client is connect but it comunicate just the few seconds.

I Have the same Problems

If you could give us access to your AP where your clients gets disconnected all the time, and we could add some debugging package then we could try to help you. Contact support@mikrotik.com

There are big problems with latency to clients with ccq less than 60 %. The are big packet lost on v 3.0. On 2.9.x It was No problem. And there are problem in compacibility with older AP like the Ovislink AP 1120. The client is connect but it comunicate just the few seconds.

I Have the same Problems

If you could give us access to your AP where your clients gets disconnected all the time, and we could add some debugging package then we could try to help you. Contact support@mikrotik.com

Ok, I sent everything, and i am waiting... thanx

This is a very interesting discussion.

I do think that hardware limits are easily exploitable with RouterOS 2.9(.50). We have Pentium 4 3400 Mhz CPU (basically maximum single-core CPU one can get) on our border router. Is has Intel PCI-E NIC's. This router has around 1100 routes in its routing table. Router has very simplistic config, BGP on global interface and OSPF on Local, almost no firewall, no nat, no queues. In fact, config is so simplistic that there theoretically can be nothing wrong with it. CPU got maxed out at around 500 mbit traffic passing through router, full duplex that is. Because of that, we were forced to add second border router and do some routing tricks do split the load.

And on the SMP issue with Linux kernel... I am running some (20+) Linux servers, most of them are either multicore or multiprocessor. Servers use linux-2.6.23-gentoo-r3 kernel (this is kernel with Gentoo Linux patchset). Some of them are high-load servers. There was not a single problem with SMP on these machines. Also Googling for some combinations of "linux" "smp" "problem" etc. does not show any meaningful results. Can you perhaps pinpoint the exact issues you meant?

Remember to also keep things in perspective... You have 500MB FDX running through a router running ospf and bgp that you paid MT about $50 for their software and you are nicely asking for more.

Consider the alternatives for what you are doing

Scott

Consider the alternatives for what you are doing

Yes, this is exactly what is going to happen. But wouldn't it be just better if SMP was working and I could stick to RouterOS there forever? (I already imagine 4 processor x 4 core machine doing 10 GE with full routing table...)
And, besides Normis said,

even the rb1000 can handle any setup you can imagine!

Forepoint,

Was your OSPF issue related to v3 also present in RC14? Has anyone else used OSPF in v3 and not had a problem. I want to upgrade my routers but OSPF operation is required. I am using RC14 without issues currently. There was nothing in the change log for ospf from RC14 to v3 so there should not have been any changes to cause problems not already in RC14.

Scott

This particular system was only installed 5 days prior to the upgrade to v3.0, it was previously running rc14 and we did not see the OSPF error prior to the upgrade, and so far have not yet seen it happen again.

We have over 45 boards (APs and CPE's) running various releases of v3, this is the only one running OSPF (temporarily replacing a x86 2.9 router after flood damage) so I have no idea if this is normal on v3 or not

Multi-Processor Support: So far, we have a large number 100+ PoweRouter 732s running Dual-Core 3gig Processors out in the field. MOST of these are running on some version of 3rc. We have had no major issues with this so far. We also have customers running 300meg full-duplex with around 41% CPU. Some firewall and mangle rules, but not a bunch.

Version3: I have experienced the PPPoE connection issue with greater than 500 connections, in our case, it was a simple fix, split the load up on multiple routers, your experience may vary. With that said, I have a number of production networks across the world that are running these with version 3rc something, and now a few already with v3.0 .. So far, we have not had any major issues with the OS crashing, or other issues.

One guy asked about a router with more than 2 cores, we are working on that as well, going up to 8 cores, but the costs get fairly high really quick. So don't know if that is going to be a production model yet or not. `

Overall, v3 new features as well as the new hardware support, far outweighs any issues still remaining. Lets get the docs updated, and more and more people using it. I'm sure there will be a few minor updates in the next few months.

As I have always say, MT great job! I look forward to supporting your software for years to come! Rock on!

I installed V3.0 to the a RB133.
CPU is't stabil.

I wrote CPU usage:
%5
%100
%7
%100
%4
%100
%12
%100
%9
%100
.
.

mkbater...

Now you already know what the reply to this will be...

Ok, I see the CPU utilization, so whats your problem? If you don't have a problem then it doesn't matter...

Scott

mkbater...

Now you already know what the reply to this will be...

Ok, I see the CPU utilization, so whats your problem? If you don't have a problem then it doesn't matter...

Scott

I didn't add any firewall rule or ext. And it don't have any traffic.
CPU is go up suddenly, and go down again allways.
I did format NAND, and I installed v3.0 again, but this is not a solution.

I will send supout to support

mkbatur,

I installed V3.0 to the a RB133.
CPU is't stabil.

I wrote CPU usage:
%5
%100
%7
%100
%4
%100
%12
%100
%9
%100


Thats an average of just over 53%...
Where is the problem??

I have theory on v3 CPU usage...
2.x calculated the cpu by showing the average over the previous 1-2 seconds, where v3 displays a realtime value at the time of updating, and as such it's going to have many momentary spikes up to 100% which is normal for any active processor.

I agree it is alarming to see, and it would be preferable for it to act the same as v2.9 but as it has been pointed out, it's not causing problems and nothing is broken, so it's not really anything to be worried about.

Normis, can you confirm my theory,? If so, any chance of reverting back to the 2.9 cpu calculation mode for easier comparison for performance differences of upgrades, not to mention ease of mind?

mkbatur,

I installed V3.0 to the a RB133.
CPU is't stabil.

I wrote CPU usage:
%5
%100
%7
%100
%4
%100
%12
%100
%9
%100


Thats an average of just over 53%...
Where is the problem??

I have the same problem although is more like 3 percent then it shows 100 % etc etc. i dont beleive that what it is showing is correct though i beleive its more like 3 percent.... UP on winbox you never see the green scale go all green which idicates 100 cpu Utilisation. ANd looking at the CPU graphs it shows an avg or 3/9 percent CPU usage, even though in the resource tab in winbox it shows 9-100 7-100 8-100... I could be wrong see what the MT team says.. -Jordan

OSPF locked up on a RB333 2 hours after installing, requireing a reboot to before routes came back online, however I forgot to make a supout...

We're seeing the same thing here. Using RB333s on our main backhaul link, we recently setup OSPF to run over the link as this is something we want to do across our entire network.

It worked flawlessly for a couple weeks, then randomly it's stops routing. All the OSPF routes are there, they arn't disabled or missing, they just don't route so there is no data being passed. ForePoint is right, a reboot solves the issue. Recently it's occuring more frequently. It has happened twice to me this morning. Next time it happens I'll save the configs/logs and send to support.

Right now those RB333s are running in NSTREME dual mode (which rocks), v3 Orc13 (if that makes sense and/or is relevant).

On your OSPF problems I had the same thing happen. COnnection tracking was turned off. I turned it on and the problem went away.

Can you confirm this behavour?

Scott

On your OSPF problems I had the same thing happen. COnnection tracking was turned off. I turned it on and the problem went away.

Can you confirm this behavour?

Scott

This problem is on 2.9.x too. I've to turn on connection tracking on all of my ospf
routers to work. If not they loose neighborhood.

Stefan

On your OSPF problems I had the same thing happen. COnnection tracking was turned off. I turned it on and the problem went away.

Can you confirm this behavour?

Scott

This problem is on 2.9.x too. I've to turn on connection tracking on all of my ospf
routers to work. If not they loose neighborhood.

Stefan

Switching on and off contrack on several routers didn't give me indication of the problem - am i missing something, Was this router RP, or ABR? what was OSPF network type?

Put rc14 and 3.0 on rb133. Works just fine. I just have no courage to put it on rb112. rb133 and 133c all seem just fine from rc14 on.
On 133 i run rip, and bgp ( interior, about 50 routes:) ) and it works nice. Peers both 3.0 on rb 333 and 2.9 and 3.0 on pc. seems to last a little longer to establish bgp with 2.9. The rest seems just fine. Nice the wireless <advanced> mode button on interface configuration. Keeps the interface simpler and easier to config, without too much information left by most to default.
I'll try ospf on these boards and see what happens with 3.0.

Urm ... just installed 3.0 final from rc13 ..

is it me, or the PPP interface menu is gone?

Was it removed from the combined final package?

check that 'ppp' package is installed

Yeah checked that. Had to install the PPP package separately.

Found it weird since it was always included in the combined OS package.

it _is_ included, I don't know how you lost him =)

On your OSPF problems I had the same thing happen. COnnection tracking was turned off. I turned it on and the problem went away.

Can you confirm this behavour?

Scott

This problem is on 2.9.x too. I've to turn on connection tracking on all of my ospf
routers to work. If not they loose neighborhood.

Stefan

Switching on and off contrack on several routers didn't give me indication of the problem - am i missing something, Was this router RP, or ABR? what was OSPF network type?

type=default.
I dont think it depends on the role of the router. It was no clear situation to give
a good error description. But enabling conntrack solved problems.

Could someone supply some links where the "multi-core issues" are described on linux?

I'm having a hard time finding any info. And curiously I see linux boxes running with 2 to 8 cores without problems. I say... *what* issues?

We ship PoweRouters with dual-core P4s and so far had no issues that I know of with using multi-cpu .

I also have big problems with OSPF. I have 3 v3 ( RB600 ) routers in my network, and about 20 v2.50 ( RB532 ), 1 main router ( x86 ). In random intervals v3 routers just stop sending the routes to the main router and I must do restart to solve this problem. When all routers was v2.50 they worked perfectly... Tried also to upgrade all routers to v.3.0, but downgraded back to 2.50 because the instability of OSPF was much higher.

On v3 i see many OSPF error messages:

1. Discarding packet - no neighboor with this source address.
2. Discarding packet - MD5 athentication failed
3. Ignoring link state acknowledgment packet - wrong peer state

also after downgrading from v3 to v2.5 wireless settings are not restored and the interfaces are disabled

We ship PoweRouters with dual-core P4s and so far had no issues that I know of with using multi-cpu .

There's no such thing as dual-core pentium 4. It's Pentium D.

Yes, I emphaze semantics.

We ship PoweRouters with dual-core P4s and so far had no issues that I know of with using multi-cpu .

There's no such thing as dual-core pentium 4. It's Pentium D.

Yes, I emphaze semantics.

Yes there is: http://www.xbitlabs.com/news/cpu/displa ... 51158.html

We ship PoweRouters with dual-core P4s and so far had no issues that I know of with using multi-cpu .

There's no such thing as dual-core pentium 4. It's Pentium D.

Yes, I emphaze semantics.

Yes there is: http://www.xbitlabs.com/news/cpu/displa ... 51158.html

No, there isn't. Your source is a 2004 rumor article. Check this out:
http://www.intel.com/products/desktop/p ... ab_pentium


Report this post

Is there a point in this dispute ?
If we really establish if there is such a thing will it make Routeros work better with multicore support ?

i believe the problem with multicpu is synchronization of threads. If one processor handles your WAN port and one processor handles your LAN port, and then another maybe handling other services, do you expect them to all just wait around for each other when packets come in? Software is lightyears behind taking advantage of current processor capabilities. Maybe allowing certain processes to be multithreaded is okay of course, but core routing and filtering probably can't span multiple procs efficiently. of course just my opinion, i could be totally wrong. it would be nice for scripting, CLI, winbox, API, route calcs, stuff that doesnt affect the core to run on alternate CPUs.

I can't see the REAL advantage of multicore at this specific moment. I mean, for specific needs, routing 100 mbit/s is just not enough to full load a high end CPU. Maybe is just my needs, but i have routerboards handling 10 mbits with ip filters and mangles, and 500/1000 mhz x86 handling 20-25 mbits with 25% cpu.
And really. For people with such demanding traffic needs that will saturate a current cpu, they should think at other solutions for managing/routing their networks. Don't mean that MT can't do it. It can, in most if not all cases. But you can't expect to route thousands of dollars networks, with a 1000 dollars machine and a max 300$ dollars router license.
And since is based on linux kernel, and multicore support is really at the begining in this too, maybe in a year or so, there will be better support for multiple cores.
And that said, think again, before asking. How exactly can you spread this kind of process that is routing/bridging among multiple cores ? Concrete examples. They really are not that many choices...... Maybe in encrypted tunnels, this will help, and all that has to do with encription. But something else ?
'nough said.

I can't see the REAL advantage of multicore at this specific moment

PPPoE/VPN termination.

p.s. Now we have better performance on free solutions, than on purchased ROS :-/
We were thinking about purchasing 6-8 more L6 licenses, but I think we'll move to FreeBSD/mpd5 solution if Mikrotik doesn't fix bugs in the next month or two...

I can't see the REAL advantage of multicore at this specific moment

PPPoE/VPN termination.

p.s. Now we have better performance on free solutions, than on purchased ROS :-/
We were thinking about purchasing 6-8 more L6 licenses, but I think we'll move to FreeBSD/mpd5 solution if Mikrotik doesn't fix bugs in the next month or two...

It's already been said encryption and tunnels.

Even without encryption, just pure termination.
600 online pppoe users without encryption, with per-user shaping and filtering and ROS is 100% loaded.
Even winbox lags and sometimes disconnects.

Even without encryption, just pure termination.
600 online pppoe users without encryption, with per-user shaping and filtering and ROS is 100% loaded.
Even winbox lags and sometimes disconnects.

You're lucky to get 600 users online.

Past 100, I always get into trouble. It's not powerfull enough, or it may need redesign . Are you using radius, and a separate radius server ? Or user manager on MT machine ?

Radius and separate radius server.
And now we have an additional service written in Perl to control PADI flood and server load via ROS API, that helped alot.
But the same amount of users can handle simple and free FreeBSD/exppp (patched user-level ppp) solution.
When we bought L6 I hoped that ROS can handle at least 1500 users on a quad core cpu...
But it's just a dream, as I can see now.

When we bought L6 I hoped that ROS can handle at least 1500 users on a quad core cpu...
But it's just a dream, as I can see now.

Yes it is.....
(Temporarily I hope.)

Just for the record...

Routing desisions can be established simultaneously while queue managment tasks take place in another process.

Happens regardless of wheather or not it is multi process, or multi processor.
Interupts and events occur which signal the processor to execute specified code....

Although there are multi processes, they can not run simultaneously on a single core, because one (or a specified number of) clock cycle equals one instruction.

To be truthful multi core can not run simultaneously either, as only one processor can access a resource (ie. memory, ethernet adapter, etc.) at any given time. That said.....
Multiple cores have an internal bus mechanism which allows them to do these things with relatively good syncronization, so it is ALMOST like having 2 processors working independantly of each other.

Never 2x the results, because the processors or cores must cooperate on resource sharing, but it is really close to 2x.