Hi,
Which one is better Routerboard vs RouterOS on x86 ? If we use RouterOS with freeradius and freeside.
I am new for this method kindly share your observations.

Routerboard hardware is much more reliable to provide a 24/7 service than PC hardware.

On x86 the cost per ether port (speaking worthy NICs, although for a BRAS two ether ports are enough) is much higher and can cost almost the same price (4 port intel gigabit nic for example) than a RB1100AHx2 alone.

Another point: suitable Routerboard (RB1100Ahx2, CCR line...) already includes an L6 license, whereas for x86 costs an additional $250.

As reliability is the paramount in these kind of setups, a routerboard wins: for a fraction of what a good x86 setup will cost you, you can get 2 routerboards and setup HA.

Then you can setup Freeradius/Freeside on an x86 machine, which will also provide you sort of a "passive" resiliency; once all users are authenticated, you can work with the freeradius system, as short downtimes won't affect already stablished pppoe sessions, only those trying to connect.

How many pppoe sessions will you have?

Routerboard hardware is much more reliable to provide a 24/7 service than PC hardware.

On x86 the cost per ether port (speaking worthy NICs, although for a BRAS two ether ports are enough) is much higher and can cost almost the same price (4 port intel gigabit nic for example) than a RB1100AHx2 alone.

Another point: suitable Routerboard (RB1100Ahx2, CCR line...) already includes an L6 license, whereas for x86 costs an additional $250.

As reliability is the paramount in these kind of setups, a routerboard wins: for a fraction of what a good x86 setup will cost you, you can get 2 routerboards and setup HA.

Then you can setup Freeradius/Freeside on an x86 machine, which will also provide you sort of a "passive" resiliency; once all users are authenticated, you can work with the freeradius system, as short downtimes won't affect already stablished pppoe sessions, only those trying to connect.

How many pppoe sessions will you have?

thanks to reply,
I have 300 pppoe users and in next 3 months that count will be 1000+. So i need a hardware that will live for next 3 years.
I am interested in VMware EXSi as in near future may be i need more then 10 GuestOS running on it. So for money saving I also want to run RouterOS on it, freeradius, freeside, testing on platypus ,dude, squid etc and many more.

If I purchased a Virtual server of Dell running 10+ Guest OS. I think i don't need more hardware in near future. So i willing to purchase it rather then purchasing a Mikrotik CCR1036 with Dell Virtual server e.g R210.

what you think of my that? please advise, is it right what I am willing to do?

I don't advise that approach. you're putting all the eggs in the same basket (PC machine) if that machine fails, you'll be in a bad situation.

If you're providing a WISP like service total network outages more than few minutes long (and even few minutes!) will make you lose customers really quick.

Also having everything on a single machine makes the setup much more complex, and the more complex, the more chances you run into scalability issues, obscure bugs and the like.

A better approach for better manageability and scalability is deploying specific services on specific devices strategically. You also have to design with high availability in mind from the ground up.

I wouldn't be so closed to the idea of buying more hardware in the future, in this business is common practice sizing for 2x in advance, then when you reach 70-80% usage, deploy a more powerful device, and reuse the previous one for other tasks, or have it as a backup.

Device's power is constantly increasing, and prices getting more competitive, so you gain nothing stocking yourself now for years in advance.

Speaking of the PPPoE service, a CCR1009 can cope with 1000 clients fine.

People is using x86 and turning to CHR for single-device application cases where the CCRs don't scale well for their actual load, like BRASes with several thousand PPPoE sessions, or multi-peered full-table BGP routers: tilera platform deployment is relatively recent and is not yet fully optimized to its multi-core potential (that will happen in ROS v7).

In line with specific-task, specific device approach, deploying e.g.

A CCR sized for your uplink bandwitdh for Provider Edge duties (Firewalling, QoS)
A CCR for PPPoE BRAS
A PC server for freeradius/freeside
A PC server for network monitoring, statistics gathering duties

Will be a better approach.

You can have freeradius and the monitoring services (cacti, smokeping, ELK) on virtualized in the same machine if you like, but IMHO a more resilient and flexible approach is having seperate machines (both for PCs, and routers) and an additional one as spare: develop a tested emergency recovery procedure plan so that in the event of a hardware failure, you can quickly have the backup device up and running.

That backup device can (and should) also be used for new deployment testing in the lab, so everything will quickly reach its ROI.

Servers sizing won't be that critical as they will be coping with one service, so you can benefit from cheaper, entry level servers; for a network of your size, an i3 machine with 8Gb RAM can cope with freeradius/CTS fine.

BTW R210 is a basic, entry level server, I highly doubt being suited for being a 10+ OS guest with all OSes being the core of your full network load.

I don't advise that approach. you're putting all the eggs in the same basket (PC machine) if that machine fails, you'll be in a bad situation.

If you're providing a WISP like service total network outages more than few minutes long (and even few minutes!) will make you lose customers really quick.

Also having everything on a single machine makes the setup much more complex, and the more complex, the more chances you run into scalability issues, obscure bugs and the like.

A better approach for better manageability and scalability is deploying specific services on specific devices strategically. You also have to design with high availability in mind from the ground up.

I wouldn't be so closed to the idea of buying more hardware in the future, in this business is common practice sizing for 2x in advance, then when you reach 70-80% usage, deploy a more powerful device, and reuse the previous one for other tasks, or have it as a backup.

Device's power is constantly increasing, and prices getting more competitive, so you gain nothing stocking yourself now for years in advance.

Speaking of the PPPoE service, a CCR1009 can cope with 1000 clients fine.

People is using x86 and turning to CHR for single-device application cases where the CCRs don't scale well for their actual load, like BRASes with several thousand PPPoE sessions, or multi-peered full-table BGP routers: tilera platform deployment is relatively recent and is not yet fully optimized to its multi-core potential (that will happen in ROS v7).

In line with specific-task, specific device approach, deploying e.g.

A CCR sized for your uplink bandwitdh for Provider Edge duties (Firewalling, QoS)
A CCR for PPPoE BRAS
A PC server for freeradius/freeside
A PC server for network monitoring, statistics gathering duties

Will be a better approach.

You can have freeradius and the monitoring services (cacti, smokeping, ELK) on virtualized in the same machine if you like, but IMHO a more resilient and flexible approach is having seperate machines (both for PCs, and routers) and an additional one as spare: develop a tested emergency recovery procedure plan so that in the event of a hardware failure, you can quickly have the backup device up and running.

That backup device can (and should) also be used for new deployment testing in the lab, so everything will quickly reach its ROI.

Servers sizing won't be that critical as they will be coping with one service, so you can benefit from cheaper, entry level servers; for a network of your size, an i3 machine with 8Gb RAM can cope with freeradius/CTS fine.

BTW R210 is a basic, entry level server, I highly doubt being suited for being a 10+ OS guest with all OSes being the core of your full network load.

Thanks,
That was nice guidance you share with us. It is very helpful
There are more thing that need your support.
1) For uplink bandwitdh for Provider Edge duties (Firewalling, QoS) at least 500mb bandwidth supported, which Mikrotik CCR device (model) is suited.
2) For PPPoE BRAS at least 500mb bandwidth supported, which Mikrotik CCR device (model) is suited.
3) For power edge duties, is there we also need load-balancing for multiple wan?or i need another one for the same!
4) For power edge duties as mainly for WAN link termination & load-balancing, I hear two competitor Mikrotik vs UBNT power edge router pro. Which one is best option?

Thanks,
That was nice guidance you share with us. It is very helpful
There are more thing that need your support.
1) For uplink bandwitdh for Provider Edge duties (Firewalling, QoS) at least 500mb bandwidth supported, which Mikrotik CCR device (model) is suited.
2) For PPPoE BRAS at least 500mb bandwidth supported, which Mikrotik CCR device (model) is suited.
3) For power edge duties, is there we also need load-balancing for multiple wan?or i need another one for the same!
4) For power edge duties as mainly for WAN link termination & load-balancing, I hear two competitor Mikrotik vs UBNT power edge router pro. Which one is best option?

1) a RB1100AHx2 will do, though I'd pick a CCR1009.

2) Any CCR.

3) I understand you mean provider edge, the same CCR from 1) will cope with load balancing fine.

4) Taking OS maturity into account, features vs pricing, and proven track record, a CCR hands down.

if you want to go with X86 do it correctly

dont do X86 to search cut cost, if you go cheap on X86 you will fail

a good implementation of X86 requires TWO well designed and powerfull machines capable of manage the work and work reliably

then deploy your services virtualized on both machines and configure it to have redundancy

of course redundancy increases the complexity of the configuration and increases the amount of work for the implementation

then connect this TWO machines to TWO manageable and stackable switches to consolidate your core network


as you see X86 is a serious business, maybe too complex for your implementation, as pukita says deployments using X86 are the result of a specific actual and real need, that is for implementations where a CCR1036 is not enough is where X86 worth it

as pukita says CCR1036 is cheap for the amount of performance and interfaces you get, to get something similar you need a X86 machine and add a manageable/stackable switch which costs itself like the CCR1036

summarizing

if you dont have a CCR1036 with a forecast of CPU usage close 75% at peak hours for this year, or some limitation of ROS on CCR are generating issues on your implementation, is advisable still not to go for X86

if you have the resources to buy the hardware, the amount of hours of work for that implementation to deploy X86 correctly and you are sure in the future a CCR will be not enough then go, you will have a nice setup

talking about CCR

i think if your budget only allow for CCR1009 go for 1009

but if your budget allows it

go for the

http://routerboard.com/CCR1036-12G-4S

this model has the best performance per dollar Ratio of the entire line

in comparison of CCR1036 with CCR1009, 1036 costs 2x the 1009 but performance is 3X time the 1009

in comparison of CCR1036 with RB1100AHX2, 1036 costs 3x the 1100 but performance is 10X times the 1100


about CCR1016 i think the performance is better than 1009 but not too stellar

i think the real advantage of 1016 over 1009 is the interface count

talking about CCR

i think if your budget only allow for CCR1009 go for 1009

but if your budget allows it

go for the

http://routerboard.com/CCR1036-12G-4S

this model has the best performance per dollar Ratio of the entire line

in comparison of CCR1036 with CCR1009, 1036 costs 2x the 1009 but performance is 3X time the 1009

in comparison of CCR1036 with RB1100AHX2, 1036 costs 3x the 1100 but performance is 10X times the 1100


about CCR1016 i think the performance is better than 1009 but not too stellar

i think the real advantage of 1016 over 1009 is the interface count

Thanks to share your experience to newbi like me, I will definitely go for CCR1036.

As someone that runs 4000+ pppoe sessions on x86 routeros I doubt that the CCR1036-12G-4S could do it.  The pppoe server code doesn't thread well (at least on x86) and other posts on the forum seem to indicate you will have one CPU absolutely at 100% while the remaining 35 cpus idle.

As far as mikrotik hardware being better quality than intel.  I wouldn't say that.  Modern Dell/HP stuff is quite reliable and you get multiple power supplies, and DC power if you want it.

Calling the CCR1036-12G-4S carrier class with a single AC power supply is kinda odd.  That would prevent us from using it right there.....