Stitching a Network All Together

[PeterO]

SmallNetBuilder is an enormously helpful resource. I've learned so much by drilling through the Charts, Reviews, How-To's, Basics, and Fora. A Huge Thank You, Tim, Doug & Crew.

I would be most grateful if someone could confirm that I've accurately synthesized SNB's recommendations and insight, as it would apply to my network-build in progress. Also, any pointers or suggestions on network topology would be great. I have yet to finalize a hardware solution to look after routing and wireless duties and I could use some guidance selecting the best hardware strategy to stitch everything together.


Some background:
I need to upgrade my small office network to better handle the expected increase in network traffic from our new backup solution. It’s a client-driven system with a NAS target. The NAS will serve exclusively as a designated backup device. No media streaming or torrent duties. Any file sharing is done peer-to-peer; we don’t run a server OS. Scheduled backups will consist of user-data folders and drive images --- both full and incremental. File archives will happen with a series of drives in rotation, separate from the NAS, and will be kept off-site. Back-up routines will happen both via wire and (regrettably) over the air.

NAS Hardware:
Intel Entry Level Storage System - SS4200-E (stock unit; and as reviewed by Craig Ellison, June 18, 2008. Noting its limitations and listed CONS, it meets are needs.)
2 disk drives; RAID 1 (mirror) (heeding, “Smart SOHO’s Don’t Do RAID”)

NAS Software:
Bundled NAS’s EMC Retrospect client software for PC/Mac


Network Clients:
1 Intel Entry Level Storage – SS4200-E; 10/100/1000 Mbps; No Jumbo Frame support
1 Desktop PC; Windows 7; 10/100/1000 Mbps
2 Network Printers: embedded print servers: 10/100 Mbps
1 Laptop; WinXP Pro; Intel 2200BG; Wireless B/G
2 Laptops; Mac OS 10.5; Wireless A/B/G/N
1 iPhone; Wireless B/G​

Network Performance Goals:
Like it’s done for so many people, my WRT54G continues to perform without fail. I prize it for its rock solid stability (v.3.1). Power cycle, what’s that? However, given that my new back-up solution includes moving an order of magnitude more data (ie. including full drive images) via copper and airwaves, it’s time to move to a Gigabit LAN with Wireless N in the mix. Here’s where I could use some guidance on network structure:


Conceptually, thanks to, “Add, Don’t Replace When Upgrading to 802.11n” I now understand the importance of putting legacy A/B/G clients on a different radio than N clients (and to use WPA2-AES encryption to maintain 11n performance). To that end, there are two methods to achieve the split: I could use a simultaneous dual-band Wireless N Gigabit router and have the 11n clients use the 5 GHz band and the 11a/b/g clients use the 2.4 GHz band.


Alternatively, I could use a two-device method --- a Wireless Router and an Access Point. Use a single-band Wireless N Gigabit router to handle the 11n clients and a Wireless G router (turned Access Point) to handle the 11a/b/g clients. As I see it with this second method, there are three flavours to this strategy.

1. Use a single-band, 2.4 GHz Wireless N Gigabit router and a single band 2.4 GHz Wireless G Access Point (ie. my reconfigured WRT54G), and ensure that each unit is fixed to non-overlapping Channels (1, 6, or 11).

2. Use a dual-band, single radio Wireless N Gigabit router and a single band 2.4 GHz Wireless G Access Point (ie. my reconfigured WRT54G). I fix the dual-band unit to use the 2.4 GHz radio frequency and ensure that each unit is fixed to a non-overlapping Channel (1, 6, or 11).

3. Use a dual-band, single radio Wireless N Gigabit router and a single band 2.4 GHz Wireless G Access Point (ie. my reconfigured WRT54G). I fix the dual-band unit to use the 5.0 GHz radio frequency, setting its channel selection to Auto-Select to enable channel-bonding. The 2.4 GHz Access Point’s channel is set to the least congested non-overlapping channel from any neighbouring wireless networks that may be present.​

Conceivably, there’s another method with a few flavours of its own: use two access points and a wired router. I’ll assume that though there may be perceivable performance benefits on a larger wired and wireless network with this method, I’m not going to see any material performance impact on what is, my small and humble network.


Outside of looking to The Charts to see which wireless routers offer the best throughput (with Uplink results weighted a little heavier, as most our laptops’ heavy WiFi traffic will come from Retrospect uploading a drive image to our NAS back-up unit), is there anything I can do to improve long-term network reliability and stability?

(Initial capital cost is always a consideration, naturally, but I have no trouble spending money on some equipment commensurate with, or a bit over the top for my needs that will have the highest chance of problem-free networking. As is the case for so many people, the personal frustration and financial penalty is too great when using a slow or worse, an unreliable network. I take one of SNB editorial points that SOHO switches and routers to a large extent are now commodities, separated only by a handful of features, personal brand affinity, price, quality level of technical support, and warranty service.)​

For instance, with the view that running any device at 100% for extended periods is never a good thing, do I materially increase a wireless router’s longevity if I relieve some of its switching duties by adding a downstream Gigabit switch and connect all my wired clients --- including the 10/100 Mbps printers and the dedicated Access Point for legacy a/b/g clients --- to the switch?

Presumably, and as I understand it given that I’m still chewing (choking) through the OSI model, by using a standalone store-and-forward switch, which reads the packet frame’s header containing (amongst other things) the respective sending and receiving client’s MAC address and forwards it accordingly (downstream if the client MAC address is connected to the switch; upstream to the wireless router’s switch for further redirection), I can reduce the wireless router’s processor workload and resulting thermal stress. Yes, I introduce a new point of failure and some microsecond latency for the extra hop between devices but I benefit from fewer potential packet collisions and mitigate saturating the wireless routers switching capability. If I’ve correctly synthesized the broad-strokes to Layer 2 data link switches in a LAN, generally speaking, is it a good practice to use a switch (unmanaged or managed) to handle all peer-to-peer clients on the same subnet? Or, are there times --- solely from a LAN perspective --- that it makes sense to connect some clients directly to the wireless router and others to a downstream switch?

(Parenthetically, I’m not sure if anybody remembers the old Schoolhouse Rock musical cartoon, “I am just a Bill”, which chronicles a legislative bill as it moves through both Congress and the executive branch in Washington D.C., but I’d love to watch a “I am just a Packet?” as a packet is moved through a LAN, then into a WAN, and beyond. Imagine the complexities of Washington. Now think of the 7-layer OSI model. Gee, I not sure which is more complicated…….)​

So, hopefully you can see that there are some things I know; some things I don’t know; some things I think I know but don’t know; and, some things I don’t know that I don’t know. All told, I may be needlessly splitting hairs given my modest network needs and limited knowledge, but it’s all in an effort to ensure maximum network health, performance, and stability; to demystify the unseen work behind those blinking LEDs and the deeply-buried menu options; to cut-through marketing hype; to avoid feeling hostage to a network structure I don’t understand; plus, it’s fun!

I’m keen to learn, so any thoughts, tips or pointers in whole or in part on my rather long but hopefully detailed enough post would be most welcome.


At the very least, thank you for reading.


Some notes:
A. Uncongested airspace: 1 neighbouring 2.4 GHz WLAN
B. Some files sent to the printer exceed 40 MB.
C. No need for VPN support
D. Internet broadband speed: ~4.5 Mbps (ADSL)

 

[thiggins]

Hi Peter,

Thanks for doing your homework!

For your wireless changeover, I recommend you stay in the 2.4 GHz band. You'll get reduced range with 5 GHz and with an uncongested airspace, there's no need to incur the extra expense of dual band. That said, given that two of your laptops already have dual-band N radios, you might opt for a single-radio dual-band router and experiment.

The things that tend to make wireless routers unreliable are bad code and thermal problems. For thermal, Gigabit switches are one culprit and processors are the other.

You can reduce thermal load on the router by using an external gigabit switch. Don't worry about any added latency, you'll never see it. You might also examine the internal photos of candidate products, to see if they use any heatsinking. Even a thermal pad on a CPU or switch can help.

When you configure a wireless router as an AP, you're essentially taking the CPU out of the equation, except when you log in to configure it. Since all traffic is on the LAN side, the CPU really isn't involved, just the radios and the switch.

As for router / AP suggestion, the D-Link DIR-655 used to be the go-to single-band router. But D-Link screwed the pooch recently with major firmware transition and their reliability and reputation has taken a hit.

I personally don't use N. An old 11g D-Link DGL-4300 serves my needs just fine. If you've had good luck with Linksys, the WRT310N is their single-band, Gigabit LAN offering.

Let us know how you make out.

 

[PeterO]

Hi Tim,

Thank you very much for your sage advice.

Thermals and Code: Roger, That. Divide thermals by separating functions.

Router, AP, and Switch:

a. WRT310n to handle Routing, and Wireless 11n clients;
b. WRT54g -- turned AP -- to handle Wireless 11b/g clients;
c. Gigabit switch to handle all Layer 2 peer-to-peer functions --- needs solid thermal design.​

I hope you don’t mind if I post a few follow-up questions based on (a), (b), and (c), and confirm my thinking.


1) Router and AP Device Separation:
With respect to separating an Access Point and Wireless Router sharing 2.4 GHz frequencies but operating in non-overlapping channels (1, 6, 11), in the Conclusion section of your "Add, Don't Replace..." article you recommend Not to place each unit next to each other in order to minimize possible interaction between them. I suspect it's difficult for you to say because on-site conditions play an enormous role in WiFi behaviour but if you were to guess, what do you think is the minimum distance in an open space I should place them apart (inches, feet, yards)?


2) 2.4 GHz and 5 GHz
If physically placing the Router & AP with the necessary device separation becomes problematic in my location (cable runs, electrical outlets, dead-spots, etc.), do you see anything wrong with this alternative, Plan B strategy:

With the considerations noted below, I can place a single ratio wireless router (5 GHz configuration) next to, or in the near vicinity of, a 2.4 GHz AP.

i). Moving to a 5 GHz spectrum yields lower range. (as you’ve noted from your large body of tests and research).

ii). It assumes that at the same range the effective uplink and downlink throughput rates for 11n clients are better at 5 GHz than they are at 2.4 GHz using a degraded signal from inter-device 2.4 GHz interference.


iii). A dual-band, single radio wireless routers costs less than introducing either Powerline adaptors or POE into the mix.​

Looking at the Wireless Charts for dual-band, single radio wireless routers, the market selection is modest but the Linksys WRT320n in the 5 GHz spectrum does well against its limited number of peers in the uplink and downlink numbers, and puts in a good showing against the more expensive dual-radio models. What do you think?


3) Firmware Code:
Funny, I’m not in software development but I appreciate the complexities of producing something in a fluid marketplace with ever-moving variables, but now with some professional experience under my belt, I recently changed my billing practices. I no longer bill by the hour; I bill by the lost hair follicle.

Perhaps born from that same experience, I try to be a little more selective over how and when that hair is lost. Sometimes it’s worth the loss; sometimes, well, not so much. Knowing that the hereto now, go-to D-Link DIR-655 wireless router of choice has been scupper’d is to say the least, an enormous heads-up. My scalp thanks you.


4. Gigabit Switch
In a decision-mix of part perceived brand reputation (solely mine), part NewEgg comments, part exterior envelop size and material choice, and part no-multi-hoop-warranty-replacement-gymnastics-policy, I picked up an 8-port Procurve 1400 series unmanaged Gigabit switch today. If I never have to factor it into a network troubleshooting session, it’s worth the price premium. …Sometimes it’s hard to count the number of times a dog doesn’t bark…


Thanks again, Tim, for your time and guidance. I’m most grateful.


Peter

 

[PeterO]

....If you've had good luck with Linksys, the WRT310N is their single-band, Gigabit LAN offering.

Let us know how you make out.

Further to your thoughts on the Linksys WRT310n...

I see in the Wireless Charts that at 2.4 GHz in a 40 MHz bandwidth, the Buffalo WZR-HP-G300NH edges out the Linksys WRT310n in average uplink throughput. It reviewed well here at SNB and unless I've missed something obvious in my research it seems that with the last firmware update (v.1.65) most of the Buffalo's connectivity issues have subsided. Therefore, given that outside of the usual small bursts of airborne email and webpage traffic, and that the bulk of the Wireless traffic through this dedicated 11n client radio with any kind of performance demands will occur during scheduled multi-Gigabit drive image Uploads to the LAN's Back-up storage unit, do you think this small advantage the Buffalo unit holds makes it on the whole a better 2.4 GHz wireless 11n building block for my network than the WRT310n?

Merci beaucoup

 

[thiggins]

For two APs in the same band, I'd separate them by 6-10 feet.

Two APs in different bands, they're probably ok to stack. But if you see any performance issues, separate them by a few feet.

The WRT320N does a decent job in 5 GHz.

The Buffalo may give you a bit more range and/or bandwidth at a given location. But it ain't gonna speed up your backups noticeably unless you're really operating at a fringe point with single-digit throughput.

 

[PeterO]

Good stuff. I'll update this post in the coming days with my results.



On a personal note, your exceptional effort with SNB and forum presence really does advance the Public Good.

With a heartfelt thank you, Peter

ps: no need to respond.

 

[PeterO]

Update

After 5 months of operation, I figure it'd be useful to post an update on my progress.


From a wired and wireless data transmission perspective, I am very pleased with network performance, stability, and my configuration supporting multiple stationary and mobile clients. Bifurcating wireless network access by dedicating a WiFi router to N clients and another unit turned Access Point for G clients works very well. It's all good news.


From a data backup perspective, it's a very different story. In short, my Intel Entry Storage SS4200 is hooped. It's been the worst type of hardware failure to troubleshoot: initial stability turns to ever-so-gradual growth of system failures -- all by way of intermittent and inconsistent symptoms. (I have lots of patience for inductive and deductive problem solving but this puzzle reached the truly rare Mad Hatter status. I put my 24 oz framing hammer on standby; ready for immediate deployment. ;-)


From the onset, Intel Tech Support has been excellent via both email and telephone. After a considerable amount of joint troubleshooting with amongst other things, crash dump files and video footage I took of the device's strange boot, power cycling, and fan behaviour -- they took it upon themselves to root out the problem not just as a customer service exercise but for their internal knowledge base -- they offered me a full refund for the unit. I would have much preferred a replacement unit, but the now discontinued unit is no longer in Intel's supply chain.


So, now back to the NAS drawing board... hammer ensconced in toolbox.

 

[thiggins]

Thanks for posting the report. We usually hear about bad vendor support. So it is nice to hear about a positive experience!