We are still on track to releasing within Q3

Not sure if anyone noticed, the image has been updated to more accurately reflect the actual product: http://routerboard.com/CCR1036-12G-4S

Alternatively this build has worked well for us for the last 2 years
http://www.reddit.com/r/mikrotik/commen ... ble_build/

We are still on track to releasing within Q3

Alternatively this build has worked well for us for the last 2 years

http://www.reddit.com/r/mikrotik/commen ... ble_build/

Our edge router is running a Supermicro board and a Core i7.

We upgraded to v5 to get support for some Intel fiber cards. The cards work but the OS is unstable. 4.17 and below is rock solid, 5.11 reboots randomly, 5.7 randomly hangs.

Delivery Time: 2-4days

I'm still curious about the pricing.

It will be very difficulty to get this beast running. It's massive parallel so algorithms have to be changed to use it's capacity. This might affect stability in the first releases...
Hope to see gigabit routing speeds with 64 Byte packets.

on 27th September 10:00AM MUM local time will start presentation you do not want to miss.

http://mum.mikrotik.com/2012/US/agenda

I thing recording will become available after the MUM.

I thing recording will become available after the MUM.

I hope so. Let us know when it does. The live feed is pretty useless right now (isn't playing)

Honzam, are you here? Come and say Hi!

ok, Normis, now show the router throughput in L3 with 20-40 rules of firewall !!!!
Proove that CCR is more powerfull that ubnt EdgeMax !! )

ok, Normis, now show the router throughput in L3 with 20-40 rules of firewall !!!!
Proove that CCR is more powerfull that ubnt EdgeMax !! )

A router of this sort isn't just going to loop packets around - How about we see what this thing can do with say 50 mangle rules, A dozen L7 filters along with a global route table?

15mill PPS is nice but my $100 switch can do that and all you've shown us so far is this fancy router being a switch

ok, Normis, now show the router throughput in L3 with 20-40 rules of firewall !!!!
Proove that CCR is more powerfull that ubnt EdgeMax !! )

We will definetly post some test results, but we can't compare, as we don't have that other device in our lab.

I hope that the version 6 have the stuck staled bgp routes bug fixed. ( since 4.x)

Hello Beccara,

15mill PPS is nice but my $100 switch can do that and all you've shown us so far is this fancy router being a switch

Surely? Please come on good old Beccara, tell us all here around also what kind of $100 Switch does Layer 3 routing, I am very interested in the brand and the performance

However this seems to have changed now and we may see some tests done

and now we can see on youtube throughput 15Mpps only in L2 !! )) But nobody say anything about L3 )

and now we can see on youtube throughput 15Mpps only in L2 !! )) But nobody say anything about L3 )

No, it is not L2, but rather fastpath, i.e. L3 forwarding in HW. The same thing Cisco does in their routers as far as I understand.

Even Ubnt can get this right. They may not be able to deliver products to our hands, but at least we know the price of the empty boxes we want

No, it is not L2, but rather fastpath, i.e. L3 forwarding in HW. The same thing Cisco does in their routers as far as I understand.

There will be a limitation to this otherwise there would be no need for MT to claim 8mill PPS normally

2 andriys - no, its not fastpath. All interfaces in bridge mode, so its only L2 ))
Fastpath in cisco not the same as mikrotik ))

Greater than 8 mpps standard forwarding
Greater than 15 mpps fastpath forwarding

Привет Харькову )

hello normis !!
Answer questions - CCR have ASIC ?? How many firewall rules i can use without throughput degradation in L3 ?? say something ))

Good ol` MT benchmarking.

A router of this sort isn't just going to loop packets around - How about we see what this thing can do with say 50 mangle rules, A dozen L7 filters along with a global route table?

15mill PPS is nice but my $100 switch can do that and all you've shown us so far is this fancy router being a switch

Good ol` MT benchmarking.

A router of this sort isn't just going to loop packets around - How about we see what this thing can do with say 50 mangle rules, A dozen L7 filters along with a global route table?

15mill PPS is nice but my $100 switch can do that and all you've shown us so far is this fancy router being a switch

And the next guy will ask for a test with 78 mangle rules. This is why we have a standard test to evaluate CPU capacity at certain conditions. You can then take your RB1100AHx2, compare results with CCR (roughly 16 times difference) and you can multiply your current setup performance with 16 to get same CCR performance. A new RouterBOARD doesn't have special CPU features to accelerate Mangle more than Routing, therefore all speed is increased by approximately 16 times (or any other amount when comparing to other device).

This is what standards are for!

Properly dividing tasks between all 36 cores is what most of our v6 development time goes to. What I wrote still stands.

Properly dividing tasks between all 36 cores is what most of our v6 development time goes to. What I wrote still stands.

MT isn't doing benchmarking right, Pick a whitepaper benchmark by a 3rd party on UBNT,HP,Cisco, Juniper - They all go into much more detail than MT do. You can't just point to one test and say it proves everything else will be x time's faster

Properly dividing tasks between all 36 cores is what most of our v6 development time goes to. What I wrote still stands.

MT isn't doing benchmarking right, Pick a whitepaper benchmark by a 3rd party on UBNT,HP,Cisco, Juniper - They all go into much more detail than MT do. You can't just point to one test and say it proves everything else will be x time's faster

Actually we follow a international standard on testing (defined by RFC2544) and using an industrial standard testing device, which can't be said by some of the ones you mention above.

Switching has nothing to do with our tests, please stop comparing the results to a switch! You can clearly see in the testing tables, which features were used. Routing/Bridging/Firewall etc.

I see a problem with this setup: the most expensive operation for conn track is to find out that a packet does not belong to
any existing connection. Testing with only 120 streams (no new connections?) does not test properly "conn tracking ON" performance.

I see a problem with this setup: the most expensive operation for conn track is to find out that a packet does not belong to
any existing connection. Testing with only 120 streams (no new connections?) does not test properly "conn tracking ON" performance.

Some link to the source of this assumption ,please.

Most traffic generators work with stateless traffic (not TCP). So each packet need to check with conntrack, how it check in the end doesn't really matters. Number of streams can be important in case of multi-core systems, so that they can be balanced on different cores (at least on x86)

While accessing a bucket is a constant time operation (hence the interest of having a hash of lists), keep in mind that the kernel has to iterate over a linked list to find a conntrack entry. So the average size of a linked list (CONNTRACK_MAX/HASHSIZE in the optimal case when the limit is reached) must not be too big.

it is know fact that CONNTRACK_MAX/HASHSIZE in RouterOS is calculated dynamically based on amount of available memory.
Also it is known fact that you need to rebuild your hash if there are entry that is new (new connection - not invalid packet).

Also in real deployments only fraction of time is spent in hash rebuilding.

So i do not see where your "problem with this setup" is - as far as i can see it is combination of easiest and closest to real application test for setups.

i do not understand why everyone is whining about it - just realize what those test shows and use this information to your advantage.
If you have small conntrack timeouts and lot of new connections - just keep in mind - that will result in bigger CPU load.
Anyway those tests are there to compare boards between themselves, to give approximate performance estimate.
So with same new connection intensity RB1100AHx2 and CCR should still relate 1:15 (with same amount of traffic your average CPU load will be ~15 times lower).

I used this tables for transition from RB100/RB500 to RB400/RB400AH and now to RB2011/RB700, and from RB450 to RB1200/RB1100AHx2, these tests helped my a lot in calculations what board i need for to upgrade, so in the future i can get necessary throughput.

Currently we use Agilent and Xena Networks equipment for our Ethernet tests.

Tests are done between CPU attached ports - ether1,ether6,ether11 on RB1100AHx2 and all 16 ports on CCR1036
Devices are configured to send traffic equally from all to all ports. (so  6combinations*20connections=120streams on RB1100AHx2 and  128combinations*1connection=128streams on CCR1036)

firewall ON - /interface bridge settings set use-ip-firewall=yes
Conntrack ON - /ip firewall connection tracking set enable=yes

Connection tracking is one of the most expensive and most commonly used features in RouterOS,
bridging currently has less performance overhead in case almost everything is disabled or not utilized.

So if we go back to our standard testing result table. you can see:

1) raw CPU power on routing and bridging by looking at 64byte packet results without conntrack
2) raw memory capability on routing and bridging by looking at 1518byte packet results without conntrack
3) CPU+memory combined performance on routing and bridging by looking at 512byte packet results without conntrack
4) then you can see how big impact on performance is from enabling connection tracking - bigger impact (in percentage) suggest that more advanced config will have bigger impact on boards performance.

this table if used properly can give you all necessary information about the performance of the board.
IF you go into specific RouterOS feature then it is not about the board, but about the implementation of that specific feature.
As i told you we are working hard to make all the features capable to use full board potential.

Hello Dobby )
I just need to aggregate some traffic, 2-6 Gbps, and drop unwanted ip or ports, and use shapers on it... And now tell me what i need ?? MX80 ?? or ASR1000 ? or CCR ? or PC + linux
I think best way is PC + linux.. On my old 8 core xeon i use about 16000 queue rules, ~3-4 Gbps traffic (4 bonding ethernet) and cpu load 5-10%
Again - if i install mikrotik on that server i get 100% cpu usage and drops.. If i set up CCR for this - i think it cant use 16000 simple queue (or mangle rules for PCQ)

Now tell me - why i need compare routers in one price gategory ??
For what i need to buy CCR ?? for 15 mpps L2 throuthput ? I better buy Dlink DGS-3120 (65 mpps wire speed + 1500 ACL rules) and it cost only 400$
What i can make with CCR that i cant make with PC or whith L2 switch ?

normis - we talk about that year or two ago )

7000 customers, 4000 online, queue with traffic priorities for each customer (4 queue per customer), two direction (in and out) ~ 4000 x 4 x 2 = 32000 queue (now i use queue only for download, so 32000/2 ~ 16000 queue)
I whant to buy Ericson Smartedge100 or Juniper MX80, but waiting for CCR with 32 cores.. And now i see that CCR not for me, its just another office switch\router

Did you try also PCQ/mangle/simple queue solution (the proper way to do it), or just created 16000 simple queues in RouterOS v5?
Looks like RouterOS v6 will be just for you - simple queue paradise.

And now i see that CCR not for me, its just another office switch\router

Again an RFC based around switching packets from port to port, What industrial device are you using? Agilent? - Everyone I quoted are alot more open about their testing than MT has been - Normally it's run by a 3rd party who detail the setup and tests run, Whats being asked isn't difficult and only MT can do it right now.

I just need to aggregate some traffic, 2-6 Gbps, and drop unwanted ip or ports, and use shapers on it... And now tell me what i need ?? MX80 ?? or ASR1000 ? or CCR ? or PC + linux

7000 customers, 4000 online, queue with traffic priorities for each customer

I think best way is PC + linux.. On my old 8 core xeon i use about 16000 queue rules, ~3-4 Gbps traffic (4 bonding ethernet) and cpu load 5-10%
Again - if i install mikrotik on that server i get 100% cpu usage and drops.. If i set up CCR for this - i think it cant use 16000 simple queue (or mangle rules for PCQ)

Now tell me - why i need compare routers in one price gategory ??

For what i need to buy CCR ?? for 15 mpps L2 throuthput ?

I better buy Dlink DGS-3120 (65 mpps wire speed + 1500 ACL rules) and it cost only 400$

What i can make with CCR that i cant make with PC or whith L2 switch ?

I whant to buy Ericson Smartedge100 or Juniper MX80, but waiting for CCR with 32 cores.. And now i see that CCR not for me, its just another office switch\router

but routers with LAG support are in normal conditions are at the price range of 5000 - 7500 € (Euro)

No this is not a good idea, please have a look over the DGS-1500-xx series they are working fine with all of my Mikrotik Routers, much better please trust me. But why you are starting it again, to compare the incomparable?

But now we see that CCR is another

You can then take your RB1100AHx2, compare results with CCR (roughly 16 times difference) and you can multiply your current setup performance with 16 to get same CCR performance. A new RouterBOARD doesn't have special CPU features to accelerate Mangle more than Routing, therefore all speed is increased by approximately 16 times (or any other amount when comparing to other device).

But now we see that CCR is another

sorry but how do you see that "now"? we have not published any performance tests or other new specifications. how do you know that it is not more powerful in all areas than your category Cisco? on what data you base your assumption?

Talking about a product without actually having it tested by yourself is guesswork.

This is just getting to be a bitch fest. Complaining about what you "think" a product will and wont do.

Someone should close this thread.

My 5 cents. If there were any good information or stats available on or around the date it was released OR made available to public

And im talking something more than a random Youtube video it would cut down the negativity, grow up , just because someone is negative dont mean you need to put them in the naughty corner and forget about them.

Im not a sales man but i know that would be an insane amount of waste in sales of you ignore those people.

Most us are technitions so, we see things from a different angle. thats why half of you get the issues in this thread that's why the other half dont.

So ill say it again if They released proper information around the release to the public was made it would have solved so many issues
I do feel like it was a massive tease, omg 36 cores holy cow OOOOMMMMMGOOOOODDD!!!

wtf... mikrotik i love your products been using them for a long time, but dam man... act professional.... I would say most of as work for company's that require us to be professional, give them what they need, and no I dont mean a comparison sheet, you find it in your best effort to give what these guys are looking for, so what if some guy wants 10 million pcq queues, i dont give two flying cats.
he just wants a bench mark on where abouts the router will start to break.... stop being argumentative gee cat wiz!

Ya all need to probably clarify that that all of us are here about the product that may make our little networks into something very awsome, stop teasing them give them the facts, they want to know what its capable of.

I would personaly like someone to answer finally the question how does the router handle an ipsec tunnel does it handle 1gbs?
i have a 2011UAS if i start a ipsec tunnel and do btest it bogs down to about 10mb/s roughly with the router doing the btest.

what can CCR do.... does it hand off any ipsec load to 1 cpu and ethernet and other functions to other cpu?

Currently we use Agilent and Xena Networks equipment for our Ethernet tests.

Tests are done between CPU attached ports - ether1,ether6,ether11 on RB1100AHx2 and all 16 ports on CCR1036
Devices are configured to send traffic equally from all to all ports. (so  6combinations*20connections=120streams on RB1100AHx2 and  128combinations*1connection=128streams on CCR1036)

firewall ON - /interface bridge settings set use-ip-firewall=yes
Conntrack ON - /ip firewall connection tracking set enable=yes

Connection tracking is one of the most expensive and most commonly used features in RouterOS,
bridging currently has less performance overhead in case almost everything is disabled or not utilized.

So if we go back to our standard testing result table. you can see:

1) raw CPU power on routing and bridging by looking at 64byte packet results without conntrack
2) raw memory capability on routing and bridging by looking at 1518byte packet results without conntrack
3) CPU+memory combined performance on routing and bridging by looking at 512byte packet results without conntrack
4) then you can see how big impact on performance is from enabling connection tracking - bigger impact (in percentage) suggest that more advanced config will have bigger impact on boards performance.

this table if used properly can give you all necessary information about the performance of the board.
IF you go into specific RouterOS feature then it is not about the board, but about the implementation of that specific feature.
As i told you we are working hard to make all the features capable to use full board potential.