How much RAM is enough in a router?

[sfx2000]

The amount of RAM you need is really dependent on how much data is going to be stored at one point and the size of the programs and the OS. So it really depends on what is needed. Nowadays 32MB is the minimum for a MIPS 24K router running the usual router stuff.

Good that you bring up MIPS - there are certain advantages to that architecture that absolutely facilitate routing - I would add IBM's Power ISA to that group as well.

Not saying that ARM or x86 is inferior, they're not, but PPC/MIPS are still valid choices...

 

[TonyH]

I guess we'll agree to disagree then.

I was taught computer basics at a different level, I guess. For what it's worth though, a cpu without ram cannot do anything. And needing less ram is not a function of the cpu, solely, it is the os and programs running. When I stated 'ram is required for an efficient and effective cpu', it implies that they are both responsible for creating something beneficial for a user. A cpu isn't effective by itself because code only runs in ram (and the code that can run inside the cpu registers and fast caches is not really usable to most users). Give a processor more ram and it will do more in return.

If you read that link to sfx2000's comments, having more ram isn't a luxury (even today), particularly for people with fast ISP speeds and many (and ever growing) devices on their 'home' networks.

Hmm, now museum piece Multics OS was a memory centered . Memory cabinet was called system control unit(SCU)
It has cache and something called associate memory to create virtual memory space used on LRU based instead of FIFO or round robin counter based and I/O mux had a scratch pad memory.

 

[L&LD]

The reason why 512 MB of RAM on a router is overkill for 90%+ of users, here's what memory usage is like on my RT-AC88U:

admin@Stargate88:/tmp/home/root# free -m
             total         used         free       shared      buffers
Mem:        515308       123256       392052            0         9304
-/+ buffers:             113952       401356
Swap:            0            0            0

I'm barely using half of what a router with 256 MB of RAM would offer.

Don't forget that heavy caching for file services is only useful if the same data is requested multiple times - a scenario that's very rarely the case when dealing with shared data (be it movies, music or other documents). Caching becomes important mostly with data that gets a lot of random access. For instance, a database. You wouldn't store a database or your Outlook PST on a network share hosted by a router, as your bottlenecks would be pretty bad.

The other scenarios that would involve a lot of random seek access are highly unlikely to be used with a router (video/audio editing for instance).

So for a router that shares a USB disk, you typically want enough caching to contain filesystem metadata. The file data itself rarely needs to be re-read a second time, so you would most likely NEGATIVELY impact performance, as that file that gets only one single access would force cache maintenance on top of the file sharing. Maintaining cache data comes at its own price, overhead-wise.

So, 512 MB might be nice for a router, but for the next couple of years, I'd say that the vast majority of users will not even fill up their 256 MB RAM with their home router. So right now, I wouldn't consider 256 vs 512 MB RAM to be a decision factor when choosing a new router. The CPU speed is far more important (and even then, only if doing anything more than providing LAN/wifi access).

Once we start getting memory intensive applications such as IPS (snort and the likes), then we'll start worrying about the memory on our router.

RMerlin, I agree with many of your points. But the option to have a router with more memory should at least be offered by manufacturers, no? Particularly as the processors get more and more powerful as they already are up to 1.7GHz dual core already.

I don't know what was running on your router when you took that snapshot, but I can't think it was anything too strenuous for it? (Features, options, USB devices and actual network use)?

http://www.snbforums.com/threads/ha...model-asus-rt-ac68a.31354/page-11#post-267950

The above link shows how an otherwise identical 'AC1900' class router is much better suited to 'normal' network loads when offered with a better performing processor and more ram. Would 128MB or 256MB ram have been sufficient to show those same increases in the link above at this time? Maybe. But as more devices are attached (and the limit I'm aiming for is 253 possible devices, since Asus supports a /24 subnet), the ram will become important for the proper functioning as intended by Asus.

Again, the best test would be an identical in every aspect router vs. one with just more (of the same type/speed) ram. Given proper test loads, that will show how important ram is.

But, I'm not holding my breathe for that kind of direct comparison anytime soon.

 

[System Error Message]

and here you were just talking about the router not being powerful enough to use all the ram . How can the manufacturer offer 2 different ram sizes when that would cost more without having a SoDIMM slot? Do we need to do a hack where we desolder the ram chip and solder a bigger one on?

 

[L&LD]

There was some concern about costs in this thread (cannot find the post anymore) of how much a router that is properly appointed (with hardware) would cost and whether it would be feasible for manufacturers to offer it.

http://www.snbforums.com/threads/up...tel-lan-8gb-ram-120gb-ssd-10-watts-257.33308/


The above link shows the wired router part of the equation coming in at about $250 for 4 LAN Intel WG82583 10/100/1000M Ethernet ports (total). That includes 8GB ram and a 120GB ssd, in addition to an Intel Celeron Processor J1900(Quad-Core 2M Cache,2 GHz, up to 2.41 GHz) processor that is many times more powerful and more efficient than any arm based pro-sumer router I know of today.

No, AsusWRT firmware is not available for this little box, but I don't think that is an hard cost today for Asus.

I also don't believe that for an additional $250 (retail, to match the price tag of the RT-AC5300) they can't add the WiFi components and the housing and still make a profit. I know why they are not doing it (money grab, of course).

I think the link above is the biggest proof yet that this level of router hardware is feasible (after all, they are making a profit on it at the $250 price level already). I am just baffled why it isn't available for purchase at all.

As the thread link above states, this little Qotom box with 4x the ram and 3x less power draw replaced a miniITX build that was 5 years old. If the ram capacity could be increased in those 5 year old boxes (stuck at 2GB max), the Qotom system might never have needed to be considered (but I would have just for the power savings alone).

When this new information is taken together with the following link below (which demonstrates that even for 'normal users', more hardware is always better), I see little reason for manufacturers to offer underpowered and under spec'd hardware for such astronomical prices like the $500 routers we're seeing for the last few months.

http://www.snbforums.com/threads/ha...model-asus-rt-ac68a.31354/page-11#post-267950


But as I stated in my previous post above, I would still be willing to pay in the $1,000 range or higher for an Asus, Synology or QNAP branded router that offers everything in one package that is stable, reliable and much, much more future proof than anything we can reasonably buy today (even from the enterprise side).

 

[L&LD]

and here you were just talking about the router not being powerful enough to use all the ram .

Where did I say that?

 

[System Error Message]

What i meant was for the consumer line, none of them have SFP or even SoDIMM slots. ASUS coming out with qualcomm's chipset and wifi planning to add m.2 but still no SFP or SoDIMM. The point is that consumers who do not have the skill to make a router from scratch wont be able to upgrade their router with new wifi or more ram. SFP is a must now that fibre optics is becoming widespread for the consumer.

Actually one of the reasons for consumer lines to have everything embedded has something to do significantly with costs in the sense of an assembly line. Read up about making PCBs. Honestly i think consumer routers are already overpriced but dont go encouraging them to make something much more expensive. As i have said in multiple posts if it is overpriced dont buy it even if you can afford it.

 

[L&LD]

What i meant was for the consumer line, none of them have SFP or even SoDIMM slots. ASUS coming out with qualcomm's chipset and wifi planning to add m.2 but still no SFP or SoDIMM. The point is that consumers who do not have the skill to make a router from scratch wont be able to upgrade their router with new wifi or more ram. SFP is a must now that fibre optics is becoming widespread for the consumer.

I already alluded to the fact that two versions of otherwise identical routers differentiated only by ram capacities is not something I'm holding my breathe for.

SFP is not a must for the home market in the near or medium future. 10GbE though, is (already).

 

[RMerlin]

But the option to have a router with more memory should at least be offered by manufacturers, no? Particularly as the processors get more and more powerful as they already are up to 1.7GHz dual core already.

It's all about cost, and market segmentation. Buying 500,000 256 MB chips is most likely less expensive than buying 250,000 256 MB chips + 250,000 512 MB chips (even assuming the two chips were the same cost), due to volume. A 50 cents difference per router amounts to 250,000$ - enough to pay the salary of a few low-level executives.

I don't know what was running on your router when you took that snapshot, but I can't think it was anything too strenuous for it? (Features, options, USB devices and actual network use)?

It's running what the router is intended to run: the core firmware, including the DPI engine, IPTraffic, dropbear, Samba, the Avahi daemon, an OpeNVPN server, etc.... LAN traffic would have zero impact on memory, as it's all switched traffic. Internet traffic amounts to 300-400 bytes per tracked connection, so even if I had a torrent with 500 peers connected to me, we'd be talking about 150 KB of RAM. Add a few more KBs per connection for buffering, you're still not going to get close to filling up 200 MB of RAM.

You are too used to the Win32 environment IMHO, where an average application eats 20-100 MB of RAM. Linux is far more optimal in its memory usage than Windows. There's a reason why a relatively busy web server needs 4-6 GB of RAM on Linux, and 16-32 GB of RAM on Windows server. A Linux web server can run with 256-512 MB of RAM. A Windows web server would need at least a couple of gigabytes of RAM.

The above link shows how an otherwise identical 'AC1900' class router is much better suited to 'normal' network loads when offered with a better performing processor and more ram.

The RAM had next to zero impact on performance there. His issue was most likely related to wifi and his configuration, not the RAM. RAM provides zero performance increase to a program, provided you still have plenty of free RAM available. That's what some of us in this thread have been trying to tell you all along. Stop thinking about a Windows PC that is starved for more RAM, and think more on how embedded Linux systems work. They are totally different beasts. Bottom line is, an Asus router with 256 MB running only the software that comes with it is not starved for RAM. A big reason for the boost to 512 MB is, quite frankly, purely marketing gimmick. Because bigger numbers make people wrongly assume that the router is better, faster. It's not.

 

[sfx2000]

I also don't believe that for an additional $250 (retail, to match the price tag of the RT-AC5300) they can't add the WiFi components and the housing and still make a profit. I know why they are not doing it (money grab, of course).

I think the link above is the biggest proof yet that this level of router hardware is feasible (after all, they are making a profit on it at the $250 price level already). I am just baffled why it isn't available for purchase at all.

Who to blame - the vendors? Or the people spending that kind of money on a consumer AP...

 

[L&LD]

It's all about cost, and market segmentation. Buying 500,000 256 MB chips is most likely less expensive than buying 250,000 256 MB chips + 250,000 512 MB chips (even assuming the two chips were the same cost), due to volume. A 50 cents difference per router amounts to 250,000$ - enough to pay the salary of a few low-level executives.



It's running what the router is intended to run: the core firmware, including the DPI engine, IPTraffic, dropbear, Samba, the Avahi daemon, an OpeNVPN server, etc.... LAN traffic would have zero impact on memory, as it's all switched traffic. Internet traffic amounts to 300-400 bytes per tracked connection, so even if I had a torrent with 500 peers connected to me, we'd be talking about 150 KB of RAM. Add a few more KBs per connection for buffering, you're still not going to get close to filling up 200 MB of RAM.

You are too used to the Win32 environment IMHO, where an average application eats 20-100 MB of RAM. Linux is far more optimal in its memory usage than Windows. There's a reason why a relatively busy web server needs 4-6 GB of RAM on Linux, and 16-32 GB of RAM on Windows server. A Linux web server can run with 256-512 MB of RAM. A Windows web server would need at least a couple of gigabytes of RAM.



The RAM had next to zero impact on performance there. His issue was most likely related to wifi and his configuration, not the RAM. RAM provides zero performance increase to a program, provided you still have plenty of free RAM available. That's what some of us in this thread have been trying to tell you all along. Stop thinking about a Windows PC that is starved for more RAM, and think more on how embedded Linux systems work. They are totally different beasts. Bottom line is, an Asus router with 256 MB running only the software that comes with it is not starved for RAM. A big reason for the boost to 512 MB is, quite frankly, purely marketing gimmick. Because bigger numbers make people wrongly assume that the router is better, faster. It's not.

Thanks RMerlin for the additional details and I agree that the ram probably didn't move the performance around too much in the link I provided (I didn't intend to imply that).

But a long time ago I too ran a version of Linux and more ram acted in exactly the same way as it did in a Windows system (more=better). In addition to better ram (faster and with tighter timings) being noticeable on any os used, Linux or Windows based.

What I don't think is appreciated here except for some (like sfx2000) is that merely having free ram is not an indication of having sufficient ram.

If programs are swapped in and out, (or, on the routers, caches and such are flushed, in addition to any other memory management the os is doing) then that is what indicates a ram shortage in the system. Showing results of free ram doesn't 'prove' anything when the router (the actual routing engine) isn't loaded to some significant degree. And I already have said that it is too bad we can't test the router when under maximum (designed) load and see what the effects of more or less ram has. As the router is based on earth based computer systems, more ram will be beneficial at that point, I'm positive.

I agree that bigger numbers in marketing make for easier sales (particularly the spontaneous or uniformed buyers). But this isn't (all of) what is happening here.

As for the cost of providing more ram? Again, not disagreeing with your position, as stated. But that is not the only option either. Instead of comparing the costs of buying 256MB or 512MB ram chips, if I was Asus I would be looking at what contract I could get for 1GB/2GB/4GB ram chips for 100M units instead, that would make more sense in the long term. And I'm sure Asus is doing that (just not at the capacity points I care about).

I am still waiting to be shown actual results of why more or less ram is needed or not. sfx2000 has sufficiently stated why more ram is needed (better than I have) for routers. The rest of this conversation has been in the 'theory' camp, imo.


I will state it once again why I want a router that is better equipped.

I want to be able to use the whole subnet (/24) that Asus allows in it's products.

I want to use all the other features it offers (except for USB attached storage, which I do not care about).

I also want to be able to use any ISP connection at full speed and lowest latency (while the router and network is at or near max load).

Right now, as I've stated before in other posts, my RT-AC68U will peg one core of the processor to 100% and even the second core will hit over 40% when I'm using a (fibre) 100d/30u connection at it's max. Yes, the ram is not maxed out during that point, but there is also nothing else happening on the router when that speed test was running either.

So, is something like what is configured in the link below what I should be aiming for?

http://www.snbforums.com/threads/up...ram-120gb-ssd-10-watts-257.33308/#post-268099


Or, maybe this?

http://www.snbforums.com/threads/ne...-dual-wan-vpn-router.32839/page-2#post-265829


Of course, I also want 10GbE ports on the thing. But I'd be satisfied to have it fast, stable and reliable with 253 users/devices attached right now.


To put this in perspective of quantum computing, more and faster ram along with a high performance processor (bit based), is equivalent to more qubits in the quantum computing world.

http://www.dwavesys.com/press-releases/d-wave-systems-breaks-1000-qubit-quantum-computing-barrier

I just wish I had the space, the money and the know how to put this type of 'compute' power to the task of protecting my 'digital front door' like the little RT-AC68U is attempting to do now.

http://www.dwavesys.com/d-wave-two-system


As the above system is estimated at between $10M and $15M, I will keep trying to get answers here.

 

[L&LD]

Who to blame - the vendors? Or the people spending that kind of money on a consumer AP...

If I had that kind of money, I would have spent it to at least test it by now, too.

So, that leaves the vendors 'to blame'.

After all, we can't buy what isn't offered.

 

[sfx2000]

It's all about cost, and market segmentation. Buying 500,000 256 MB chips is most likely less expensive than buying 250,000 256 MB chips + 250,000 512 MB chips (even assuming the two chips were the same cost), due to volume. A 50 cents difference per router amounts to 250,000$ - enough to pay the salary of a few low-level executives.

Or one overpaid sales exec with a big expense account

It's running what the router is intended to run: the core firmware, including the DPI engine, IPTraffic, dropbear, Samba, the Avahi daemon, an OpeNVPN server, etc.... LAN traffic would have zero impact on memory, as it's all switched traffic. Internet traffic amounts to 300-400 bytes per tracked connection, so even if I had a torrent with 500 peers connected to me, we'd be talking about 150 KB of RAM. Add a few more KBs per connection for buffering, you're still not going to get close to filling up 200 MB of RAM.

And if you have 512MB of RAM, and using less than 256MB, that's a pretty poor design - seems like AsusWRT might have some limits that someone has overlooked - a Linux system is properly balanced when all RAM is in use, and we're not hitting swap in an embedded platform.

You are too used to the Win32 environment IMHO, where an average application eats 20-100 MB of RAM. Linux is far more optimal in its memory usage than Windows. There's a reason why a relatively busy web server needs 4-6 GB of RAM on Linux, and 16-32 GB of RAM on Windows server. A Linux web server can run with 256-512 MB of RAM. A Windows web server would need at least a couple of gigabytes of RAM.

Let's not go down that rathole of Win32 vs. Linux - they're very different in how they are architected, and how they use memory. But to keep discussion short, Linux is much more efficient on how it uses and manages memory.

I've never stated that more RAM will be better speed (that's one peg of performance), but more RAM is more capacity - more connections for all clients on the LAN behind that router.

We're right at the cusp of seeing the number of connected devices expand by a logarithmic order of magnitude.. within the next 12-18 months, so it's better to get ahead of the problem now, considering the life cycle of consumer Router/AP's is 24-36 months.

An OEM's need to start thinking about it now - we've seen some very good mileage out of what is basically the GPL drop from Linksys on the WRT54G, but between security, stability, and performance, we're running out of time...

 

[sfx2000]

If I had that kind of money, I would have spent it to at least test it by now, too.

So, that leaves the vendors 'to blame'.

After all, we can't buy what isn't offered.

I'll spin that around - the Vendors are going to charge what the market bears - bigger numbers, better performance, right? And people buy that - that and check boxes on a feature sheet...

So no, I don't 'blame' the vendors - if they can put $100-120 of "stuff" into something they can sell for $400-500, why not?

PT Barnum was a master of marketing after all...

 

[L&LD]

I'll spin that around - the Vendors are going to charge what the market bears - bigger numbers, better performance, right? And people buy that - that and check boxes on a feature sheet...

So no, I don't 'blame' the vendors - if they can put $100-120 of "stuff" into something they can sell for $400-500, why not?

PT Barnum was a master of marketing after all...

There is no way they are putting $100 into something like an RT-AC5300, imo. Can't convince me of that (I did work in retail in the distant past).

I'll agree that as long at they can make it and (keep) selling it, they should. But that doesn't stop them from offering an even better (hardware) option for the remaining few hundred thousand that would need it.

A company that rests solely on it's past accomplishments is going nowhere but down, until (and if) they change. Apple is the poster child for that fact now.

 

[sfx2000]

We're right at the cusp of seeing the number of connected devices expand by a logarithmic order of magnitude.. within the next 12-18 months, so it's better to get ahead of the problem now, considering the life cycle of consumer Router/AP's is 24-36 months.

An OEM's need to start thinking about it now - we've seen some very good mileage out of what is basically the GPL drop from Linksys on the WRT54G, but between security, stability, and performance, we're running out of time...

And the carriers are seeing this - and they're pushing the vendors in their ecosystem (Arris, Netgear, Actiontec, etc) to get ahead of the problem now... they want to be able to push additional services for their customers, and their customers in turn will either buy them, or substitute them with over the top providers.

I've been dancing around this, but I'm sharing some very valuable insight here - for the 3rd party vendors, they need to keep up with the equipment that the carriers are putting into the home these days - capacity, stability, and performance wise - the recent crop of carrier gear is actually not that bad - and much better than the tribal consensus is...

 

[sfx2000]

There is no way they are putting $100 into something like an RT-AC5300, imo. Can't convince me of that (I did work in retail in the distant past).

I did a BOM analysis post a while back on an AC5300 class device - and if I recall, it was roughly $127 FOB Shenzen...

 

[sfx2000]

Found it -- http://www.snbforums.com/threads/new-ac-router-which-will-be-supported-longest.31493/#post-250410

Numbers have changed a little bit due to spot-markets...

 

[L&LD]

Found it -- http://www.snbforums.com/threads/new-ac-router-which-will-be-supported-longest.31493/#post-250410

Numbers have changed a little bit due to spot-markets...

Your analysis and calculations are great. But Asus is buying on a different playing field from anything we can get access to. No way 'spot market' comes into play in their costs (unless it's in the contract that they get that price and lower, if it ever dips below their negotiated price).

Even then, if they're actually making $400 per router or so, anything over and above 'reasonable' (and this isn't) will come back and bite them in the behind. Karma is blind.

 

[sfx2000]

Do need to clarify a point here - as I don't want to confuse folks...

For a given arch - more or less RAM will not make things faster, that's a capacity thing.

That being said - faster memory, more efficient memory controllers, wider paths to memory - this is all about performance.

Most consumer gear is engineered to a price point - and that is that - less than 500Mbps on the WAN side, and less that 50 connected clients - you're probably ok... wouldn't notice the difference until one puts things under the knife... once you get above those items, then things get interesting...

And with IoT coming on with capacity concerns - more RAM is better... add faster pipes, then the mem speed/width comes into play...