RB850Gx2 vs RB450Gx4

`

:laughing:

– Nathan

I’m considering the local provider’s gigabit GPON offering, which comes with an ONT with AC wifi, but I Want to use the PPPoE pass-through option. Would I be able to saturate Gbit wtih an RB450Gx4 and PPPoE using NAT and around 10 effective FW rules?

So, now that you have ā€œcorruptedā€ ( :laughing: ) the 400-series with a non-MIPSBE CPU, you are faced with the challenge of efficiently communicating this on the ROS download page. Right now, if I am the proud owner of a new RB450Gx4, and I go to this page, I’m being told I need to download the MIPSBE package.

Despite what everyone says, CPU archs really did by-and-large stick with particular series (whether by coincidence or design), and that made this kind of thing a lot simpler. Until 1100AHx4 was released, the download page showed ā€œRB1xxxā€ in the PPC row (I checked it, on the WayBack Machine / archive.org). Now because of the ARM-arch AHx4, you are forced to list every single PPC 1xxx-series model separately, and because of this you missed some (specifically, RB1000 and RB1100Hx2 do not appear in the list on the download page). But the number of RB4xx models that are MIPSBE are too many to list separately like that. Maybe this changes to ā€œRB4xx (except RB450Gx4)ā€? And then keep adding exceptions until the list is 100 miles long?? This is going to get real convoluted real fast… Probably need a completely new download page layout model and way of labeling the downloads to accommodate this kind of thing.

I know, I know…beating a dead horse is my hobby-horse. :sunglasses:
mipsbe-models.png
– Nathan

Has really no one attempted using Gbit PPPoE on this device yet?

i think for that scenario the best choice is rb1100ahx4, it has the highest single core performance of all product line, and the best single tunnel ipsec accelerated performance

remember rb450gx4 uses 716mhz 4 core CPU, at the instant you saturate one core you can get into performance problems

rb450gx4 716mhz x core arm cortex a7
rb1100ahx4 1400mhz arm cortex a15

2x performance advantage by core clock
2x performance advantage by core architecture

almost total 4x the performance x core advantage for rb1100ahx4

@chechito, chanks for the insight, though comparing the devices in itself doesn’t tell much. Obviously the RB1100 series is way faster. But many small, cheap routers are capable of what I ask and I think for MT to stay competitive in that price range they should be able to handle that, too. There’s also ā€œPPPoE fast pathā€, supposedly helping PPPoE performance, but I’m not sure if that’s relevant here or not.

the fact is

mikrotik can do very high speeds on small devices being very competitive, like another vendors or better

but

mikrotik have a very wide spectrum of functionalities beyond competition that some day maybe you wan to use, like heavy firewall and QoS and traffic management


you can go with a hap ac2 for 1gbps if you are 100% sure dont need heavy features

i am warning you about that because is very frequent to see at the forum complains about the change in performance between for example:

fast-track nat performance VS queueing performance

because that i recommend rb1100ahx4 because 1gbps is serious bandwidth when talking about some heavy features in place

thats the difference, you can do basic NAT at 1gbps with cheap devices, more features will require more hardware

@chechito: I stated my needs. I don’t need a $300 router. Believe me, I don’t mix up heavy queues with some NAT or filter rules. I also separate my APs and gateway, though HAP AC^2 and RB450Gx4 use a similar CPU. After reading posts on other forums and also here I concluded that the RB450Gx4 would be more than capable for my needs so it’s a worthy contender to replace my old trusty RB450G. If I ever get the gbit uplink (I’m pondering since the price difference is so small, compared to slower ones) I’ll definitely report back.

good

enjoy that rb450gx4, that“s a nice device :sunglasses:

As far as I can tell, from a processing power perspective, the two SoCs that hAP ac^2 and RB450Gx4 use (IPQ4018 and IPQ4019) are basically identical. I suspect the only reason that 450Gx4 uses the 4019 is for the SD card interface.

Unless you have specific needs that the Gx4 meets (like microSD slot and huge RAM and NAND storage), and you are mostly interested in CPU power for forwarding packets, I would probably save the $30 (+ whatever you need to spend on a case and power supply!) and just get a hAP ac^2 instead. Both have same CPU ā€œoomphā€, and although you lose RAM + NAND + microSD, you gain wireless + USB + it comes with a case and PS.

Without MetaROUTER support on ARMv7, I’d guess the primary appeal of the Gx4 is for people who want the most cost-effective Dude server device.

– Nathan

Isn’t 750Gr3 more cost effective as dude server than 450gx4?

`
You’re right, I forgot that MMIPS can run Dude server. However, you will need to purchase and add storage for this in the form of a microSD card, whereas the Gx4 has 512MiB of NAND flash already. (Likely you could get a 2GiB microSD for very cheap, though.)

(And speaking of…why the **** is there no Dude server package for PPC, when it exists for MMIPS? Ridiculous. Any multicore PPC RouterBoard would be more than up to the task. But I digress…)

Perhaps I should have said that the 450Gx4 looks to be the most cost-effective hardware option if you want to run Dude at anything close to scale. Our x86 Dude server with 2,700 devices constantly has half-a-gigabyte of RAM consumed at all times, and the storage required for this seems to be around 700MB.

I don’t know how well the CPU on the Gx4 would deal with our Dude database, and the storage would be not quite enough (though could also be augmented with microSD), but at least it would have enough RAM.

– Nathan

None with clear mind would destroy internal nand by dude database. You always need external flash for that if you would like the device boot after a year of running the dude.

ohh yeah bad idea to use integrated and expensive storage for the dude

better option and cost effective:

hAP ac2 + USB drive

10 us costly than rb750gr3 but with 2 wifi included and far more cpu power to move traffic in spite of the dude cpu usage,

True.

`
The CPU may or may not be better than the multicore MIPS (I haven’t looked at the test results, and also the Dude may exercise the CPU differently than routing/packet forwarding would). But even if the quad-ARM is objectively better, Dude also likes RAM, and even though some lucky hAP ac2s shipped with 256MB, on paper it’s only supposed to have 128, while the hEX Gr3 is spec’d for 256. If you buy a hAP ac2, you are not guaranteed to get more than 128MB of RAM.

Gx4 still wins when it comes to RAM + CPU power (no compromise on either), and has microSD slot. So it really just depends on what you use the Dude for and how big your database is. At $99 the Gx4 is not much more expensive than either the hEX or the hAP ac2, so if the primary reason you are buying a RB is for Dude, then it seems like that extra $30-40 is well-spent.

– Nathan

I found this page on the HAP AC2 the other day. I thought it’s relevant because the CPU is almost the same, barring wlan capabilities in the RB450Gx4. It’s mostly throughput tests (including PPPoE over Gbit), in Russian but the screen shots should speak for themselves.

keep in mind we are talking about very little CPU cores

dont expect to move a heavy instance of the dude on a router, even on a rb1100ahx4 (4x the performance per core) you have to keep an eye

forum is full of failures because people not sizing their dude implementation properly

the dude on a router is not a substitute for a real x86 server

the dude on a router is a very useful feature for small, near, and distributed monitoring

but forget about a big database or many objects

please dont use integrated nand flash to host the dude database, is very write intensive and can kill flash prematurely

VPS for Dude posts moved to separate topic.
http://forum.mikrotik.com/t/a-vps-to-run-dude/120794/1

The test results for Ethernet test result for the two models seem to be consistent in that RB450GX4 is faster than the older RB850GX2.
But IPsec throughput test published seems to indicate the older RB850GX2 model performing better than the new RB450GX2, by quite a bit too.
I wonder why that is? Is it a mistake? Or difference in test methodology?

I think it’s a hardware capability

rb850gx2 is based on Power PC SOC, similar to rb1100ahx2 in features and performance of ipsec, except clock speed

maybe this powerPC SOC was more focused on enterprise market than the SOC of rb450gx4 more focused on mainstream (this SOC includes wifi but not enabled on rb450gx4)