wkdshadow
Posts: 129
Joined: 2/6/2008
Status: offline
|
It's true that latency and bandwidth are two seperate things. Increasing bandwidth does not reduce latency. http://www.stuartcheshire.org/rants/Latency.html explains things well.
Server load plays a role, but it's not that much for a forum. Server load can be mitigated. For example, if collarme was moved to PHP, lighttpd, and mysql it'd be much faster. From a quick nmap, I know it's on Server 2003 w/ IIS6. That makes me assume it's got asp and mssql behind it(eww). Find my post in http://www.collarchat.com/m_1645172/mpage_2/tm.htm# for why this is retarded in servers.
But, that's not the point. Comcast is bitching about 10% of people using 90% of their bandwidth - well, they are selling it as "unlimited", and they're overselling capacity past safe limits. ReadyBoost screws them even more, because it gives a sliding-scale QoS profile to things, which allows them to reduce load a bit doing QoS, but makes peer to peer traffic more efficient by design.
Basically, readyboost polls active connection tables and averages the speed after 10 seconds or so. You effectively have 10 seconds or so of uncapped traffic for each new connection, longer in high load areas where they've got 600 homes to a single CMTS. Well, when someone pops open a peer to peer application that's popping open half-open connections like crazy, each connection gets a high priority QoS profile by default, before getting slapped down with the bandwidth cap. Rather than actually improve network conditions with this, they make things worse, because now people in your neighborhood don't just take up 600-900kb/s of bandwidth, if a download accelerator is used and properly configured they're running at ~1200-2400kb/s(in my experience in Utah).
They just made things worse by favoring the 10%. So, they introduce Sandvine, which starts manipulating window sizes for TCP connections that've been open for longer than a few minutes. But, this breaks shit - not just bittorrent. And really, it doesn't break bittorrent anymore, it was a temporary setback. So now, SSH(which is just an encrypted version of telnet) dies randomly, along with anything else over TCP.
The arguement that businesses using VPNs should have "business" connections is retarded. Just because a VPN is in use doesn't mean it's using lots of bandwidth. If there's nothing going over the network segment, the VPN sits there idly. And what's it matter anyways? It's the same piece of coax for both plans, and most business plans are slower than their consumer counterparts anyways. Buying a business plan is supposed to offer you protection from network outtages, but your modem will go off with the rest of the neighborhoods when the CMTS gos on the fritz, or because a geek that knows his way around TCP/IP gets his download accelerator *just* right and he's getting shit off usenet faster than the CMTS can handle it - using both network resources and CPU/memory resources at the headend. If they had seperate coax for business class connections I could see why, but they don't. It's a shared infastructure.
As for the files, ~2mb isn't big. I'll assume you mean megabytes, not megabits. That's another funny thing... the difference inbetween bits and bytes. Most people don't know the difference, but here it is: A Byte is(now a days) 8 bits.
So, a 2mB file(2048kB) on a ~12MB(~12288kbps, or ~1536kB/s) connection will take you about 1.333333 seconds to download a 2mB flash file. That's not too bad considering in 98 a T1 was still considered the shit.
To illustrate the reason why a business class connection is generally counterintuitive now-a-days, that same file that took 1.3 seconds to download over a cable modem will take 10.416667 seconds on a T1, because a T1 is only 1.536mbps anyways, compared to the 12mbps of Cox here in Phoenix. You'll also pay several hundred dollars a month to be gaurenteed those 24 DS0 lines(standard telephone line) coming into your building will be up, and lately you aren't even gaurenteed that connection won't go out if your only broadband option is DSL in the area, because it's over the same telephone trunking lines. If the DSL modems go out and it's not a problem with the DSLAM, but the lines themselves, you're shit outta luck. Someone hit a telephone pole, and both lines are out. The difference is the T1 gos over bare copper the whole way, the DSLAM gos to a fiber optic network with more speed and higher reliability. And if the service does go down, as long as it's less than 24 hours you can't do anything about it anyways, because if you've ever signed a service contract for a DS1(T1) line, you'll notice in fine print it's not gaurenteed always on, it's gaurenteed to not be down for more than 24 hours at a given time. Same contractual agreement you get with a consumer connection.
Now, like I said, there used to be a point. A DS1 used to be the shit back when everyone was on dialup. A DS3 was even more mind blowing. But this isn't 1998, and a "business class" line makes no financial, or technical sense when put into perspective now-a-days. It's just as failure prone.
You can't generally get a real fiber optic line to your house anyways to get a true reliable business connection. FIOS is fucking funny as hell to me, because they might as well put a few 24 port switches in a telecom cabinet and run cat5e/cat6 to your house. They'd be spending alot less money. The fiber being put into the ground right now can't even do full 100mbps, because it's not glass - it's cheap plastic shit. Cat6 is cheaper too, and can do gige runs to your house - not that they'd ever give you 1000mbps of WAN bandwidth, but they'd be able to run alot more bundled services to your home over ethernet vs the pseudo high speed shit they're putting in now. If they were really planning to increase network capacity, they'd do 10gige over cat6 - you can use cat6 for 10gige if your runs are under ~170 feet. For longer runs, they can use 802.3an for up to 330 feet, which is again copper.
But, it's all about politics anyway... why would they ever want to deploy network infastructure that's future proof? It'd hurt their margins and tax breaks.
PS: Copper has lower latency than fiber.
PPS: Why aren't we doing jumbo frames over WAN yet? Put simply The MTU(maximum transmission unit) of the original ethernet standard was 1500 bytes, that is to say that any packet of data contains a maximum of 1500 bytes of data, for 1522 bytes maximum per dataframe. If providers were actually interested in making their networks more efficient, they'd move to jumbo frames - MTU set to 9000 -, or superjumbo frames at anywhere between 15000 and 64000.
< Message edited by wkdshadow -- 2/27/2008 3:37:28 AM >
|
Comcast jammed the FCC hearing~!!! - 
Profile
RE: Comcast jammed the FCC hearing~!!! - 
New Messages
No New Messages
Hot Topic w/ New Messages
Hot Topic w/o New Messages
Locked w/ New Messages
Locked w/o New Messages
Post New Thread