Tuesday, February 2, 2010

Wireless : State of the Art (Or how Wireless I learned to love ethernet)

I'm moving from my old site underabundant.com
July 3rd 2009

This is a response to Rob Landley July 2nd, 2009.

Networking requirements are like computers, they need a killer-app to drive demand for faster speeds and capacity. I work at a University that uses VOIP technology. This has really driven us to standardize our network (cabling and speed minimums) at pretty high bandwidths in order to maintain voice quality. Old cabling (cat5) is generally inadequate for many applications and right now we are standardized on Cat6 (a higher grade variety which I won't go into) for all of our new installations. But why? We generally use only 100 Mb networking, and cat5e can actually support gigabit anyways so that seems a little insane. Why would you pull cable that can handle 10G to offices that won't likely see true gigabit for years?

The reason is that you have to cable everything on the assumption that it will last for 10-15 years. So we pay extra now assuming the future will bring technologies that require this cabling so we don't have to spend tens of thousands of dollars to redo everything. In our current position we are stuck with several buildings which weren't really cabled with the future in mind, who's connections are cat5 in ulgy little ethernet panels that were designed to fit into mounting brackets for analog lines! But either way these places were cabled correctly, it's just been 10 years, and they were originally cabled with 10Mbit hubs in mind. This didn't really even concern us until 5 years ago when we actually cared more about the client's speed and not just whether they were connected.

Secondly, wireless really generally sucks beyond having one in your apartment. Even two in a house starts sucking pretty hard pretty fast. It really does. User-based logins, proper management of radio channels, QOS and managing coverage holes are more than challenges, they're impossibilities. The wireless standards started from the idea that you might have one AP in an area and expanded from the bottom up. So now there are a lot of people trying to bolt on their own solutions to a really short-sighted development of standards.

As of last year we had old crummy stand alone APs and the system was extremely complex and managing it was a terrible pain. Even now we "modernized" to Cisco's wireless controllers and I have to say I am not impressed. The controllers ability to see issues or even record network activity is so limited that when people ask if x user was in y location, we just don't know. These are basic requirements that just haven't been met. Not to mention the number of bugs in the system, wireless is just nowhere near being an essential service that we can hope to have functional with any reliability.

Note: There is a location appliance by Cisco, but it's shit, it requires every AP be with 10-20 feet of each other ($$$), or to have little sensor devices installed everywhere in the ceilings ($$$+$). None of Cisco's ideas really appear to scale when you have more than a floor of a building. In fact they seem downright retarded in the long-run. I find that generally at the University, projects operate on whether management can be convinced that's it's easy so that they can get carte blanche to shovel as much equipment and services onto the PO.

Also, dense wireless is kludgy, you can have an AP a few feet from you, but connect to a different AP in a building across the street for no reason. Or stay connected to an AP 100 feet away as you pass 3-4 closer ones while on the move. Once you have more than one AP on the same SSID, you get problems. As an aside Meru networks has a supposedly standards-compliant fix for this. They have controllers that decide where clients connect instead of the clients deciding. This is good because clients are usually stupid and selfish and the controllers can more easily negotiate the users and tell them where and when to connect. But they are not number 1 and they are the only ones to have this solution. I hope that it gets out though.

For wireless you require site surveys to ensure coverage, which are pricey and granularly anal, and you might be surprised how easily a moderately dense area can max out an AP so of course lots of areas need two somehow. The max is really something like 10-15 people per radio for a regular network connection. God forbid they were all VOIP phones or doing streaming. Not to mention the fact that you still have to purchase cabling for these devices anyway. I've lost count the number of times my boss said we can just throw up an AP in an area that can't be cabled. If it can't be cabled, it can't be wireless either unless you want to hire electricians for a power plug and setup a wireless bridge to provide the network (you don't). Wireless is full of possibilities, but it's really not much of a solution yet, regardless of how people see it.

My bosses dream of a captive portal for all wireless users but I don't think they realize there really isn't a _generally_ secure medium for that yet either. There really isn't anything at all for securing a connection through a HTTP page. So everyone putters along with their complex enterprise configs and all our service areas are loaded up with. Everything needs to be patched in order to work because all the standards are designed assuming it's in your living room, and all the companies are just expanding on that and slowing adding the convenience of enterprise solutions. Just recently we turned on wireless-N capabilities and the new configuration (thank god 90% of the 20,000 students didn't need to change it) caused a number of "legacy" cards and platforms to no longer work.

I have seen one university that just sets a trivial static WPA password and then forced people to log on using captive portal. This is kludgy at best because you still have to get the word out for that password, so the efficiency of a captive portal is diffused somewhat. It's a hack!

Also, wireless is expensive and doesn't cover a majority of the clients on our network. For 1500$ per AP, one AP every 30-40 feet depending on obstructions and bandwidth needs and you might realize how much cheaper it might be for 100$ ethernet jacks and a well managed telecom room. Even with some wireless coverage in the same area. Not to mention services like copiers, faxes and most desktops, ethernet has a lot of time left.

So I think you can understand a little better why mandates for wired cabling is still happening. Especially when it comes to building construction. My boss told me once that our jobs will be defunct in a few years because of wireless, but it's just not there yet, not even close. Wireless will become the new switches when they become practical and wide-spread, but as of now I think Apple is the only computer maker that puts wireless N cards in everything (not just as an extra, but everywhere).


Post a Comment

Subscribe to Post Comments [Atom]

<< Home