Akami Web Visualizations
June 7, 2007 3:20 PM Subscribe
Akamai's Internet Visualizations. Akamai is a major mirroring and caching service which serves up a large chunk of all internet traffic. They are now sharing some pretty visualizations based on their data which used to be customer only. News. Music. Retail. Real-time Web Monitor . Network Performance Comparison.
Visualizing Akami.
More fun to watch than your computer's CPU meter, certainly.
posted by davejay at 3:31 PM on June 7, 2007
posted by davejay at 3:31 PM on June 7, 2007
Customer only, but they show it on big plasma screens in their NOC that I walk past every day. Yow.
posted by mkb at 4:15 PM on June 7, 2007
posted by mkb at 4:15 PM on June 7, 2007
which used to be customer only.
They were showing the marketing material to existing customers only?
I love how they illustrate their lower latency by drawing a straight line across the atlantic vs the "longer" curved one used by the "public internet".
(which reminds of some whiteboard scribbles I saw at Google a few years ago -- step 1: dig tunnel through center of earth to reduce latency, step 2: replicate earth to increase throughput)
posted by you at 4:20 PM on June 7, 2007 [2 favorites]
They were showing the marketing material to existing customers only?
I love how they illustrate their lower latency by drawing a straight line across the atlantic vs the "longer" curved one used by the "public internet".
(which reminds of some whiteboard scribbles I saw at Google a few years ago -- step 1: dig tunnel through center of earth to reduce latency, step 2: replicate earth to increase throughput)
posted by you at 4:20 PM on June 7, 2007 [2 favorites]
HTTP Hits per Second: 2,517,750
Akamai handles 20% of the world's total Web traffic
Amazing.
posted by stbalbach at 4:24 PM on June 7, 2007
Akamai handles 20% of the world's total Web traffic
Amazing.
posted by stbalbach at 4:24 PM on June 7, 2007
According to Wiki, the name Akamai is allegedly a Hawaiian word meaning "smart", but I've always assumed that it was a wordplay joke. Akamai backwards is "iamaka", or as I read it "I am A.K.A." ("also known as"). Since they started as a DNS lookup system before going into the caching and reverse proxy business that... kind of makes sense, yah, since DNS is like an A.K.A. for IPs. :) Also, at least a few years ago their customer control panel portal was found at i.am.akamai.com ("I am Akamai") which is a palindrome (er, without the .com of course).
Since mkb apparently works there, I assume he can confirm if I'm on the right track?
That said, they have a neat system- it's essentially putting a few servers in almost every ISP and datacenter around the world; these systems are then managed in a widely distributed and automated fashion, and it gives them some kind of presence in the closest datacenter to most everyone in the world. I know in the search engine world, they are heavily used as in the race to be "quick" with results the 400-600ms a customer spends on RTT for each call to a page element to an overseas datacenter can be shaved to < 100ms by a) caching and b) establishing a longer-term session to a nearby server, which can remove the creation/teardown time of the tcp connection. their technology is pretty sweet, but also crazy expensive and prone to problems,
posted by hincandenza at 5:05 PM on June 7, 2007
Since mkb apparently works there, I assume he can confirm if I'm on the right track?
That said, they have a neat system- it's essentially putting a few servers in almost every ISP and datacenter around the world; these systems are then managed in a widely distributed and automated fashion, and it gives them some kind of presence in the closest datacenter to most everyone in the world. I know in the search engine world, they are heavily used as in the race to be "quick" with results the 400-600ms a customer spends on RTT for each call to a page element to an overseas datacenter can be shaved to < 100ms by a) caching and b) establishing a longer-term session to a nearby server, which can remove the creation/teardown time of the tcp connection. their technology is pretty sweet, but also crazy expensive and prone to problems,
posted by hincandenza at 5:05 PM on June 7, 2007
You can't really trust what they're showing you about 'routes on the internet versus Akamai' either. Atlanta to Tokyo through Africa? That seems pretty unlikely. And Atlanta to New York shows as routing through Venezuela or Colombia, which is completely ridiculous.
My takeaway: these are lovely maps, beautifully presented. They are also blatant lies.
posted by Malor at 5:26 PM on June 7, 2007
My takeaway: these are lovely maps, beautifully presented. They are also blatant lies.
posted by Malor at 5:26 PM on June 7, 2007
interesting the number of attacks on Venezuela's network. ~6X more than China even. Wonder who's behind that?
posted by twjordan at 5:33 PM on June 7, 2007
posted by twjordan at 5:33 PM on June 7, 2007
According to Wiki, the name Akamai is allegedly a Hawaiian word meaning "smart", but I've always assumed that it was a wordplay joke.
Well, there is a Hawaiian word, akamai, that means smart, or clever. I always figured that was what they meant. The palindromic interpretation sounds cool, but it feels like overthinking the issue to me. Could be wrong, though.
posted by rtha at 6:05 PM on June 7, 2007
Well, there is a Hawaiian word, akamai, that means smart, or clever. I always figured that was what they meant. The palindromic interpretation sounds cool, but it feels like overthinking the issue to me. Could be wrong, though.
posted by rtha at 6:05 PM on June 7, 2007
it's essentially putting a few servers
I actually worked for these guys a couple of summers ago, maintaining and upgrading their servers in several data centers here in northern Virginia, and I can tell you, it's more than "a few." In the installations I dealt with, there were hundreds of servers. Rack after rack after rack of them.
posted by Dave Faris at 6:36 PM on June 7, 2007
I actually worked for these guys a couple of summers ago, maintaining and upgrading their servers in several data centers here in northern Virginia, and I can tell you, it's more than "a few." In the installations I dealt with, there were hundreds of servers. Rack after rack after rack of them.
posted by Dave Faris at 6:36 PM on June 7, 2007
Whenever I come across any mention of Akamai I can't help but think of and remember Daniel "Danny" Lewin -- co-founder (and CTO) of the company who was killed aboard American Airlines Flight 11 during the September 11, 2001 attacks.
posted by ericb at 10:08 PM on June 7, 2007
posted by ericb at 10:08 PM on June 7, 2007
I loves me some infographics. Thanks!
posted by Civil_Disobedient at 4:29 AM on June 8, 2007
posted by Civil_Disobedient at 4:29 AM on June 8, 2007
I worked for Akamai's NOCC for about two years (and yes, it has two "C"s. Marketing calls it the "Network Operations Command Center," in order to differentiate from a regular NOC). Basically, the visualizations are pretty pictures, rather than anything that is used by the engineers and technicians staffing the place, though occasionally having a look at overall throughput or other general trends might give you a clue about what's going on ("Why did our throughput just increase by 30%?" "Oh, that's because BigCompanyX just released a new patch.").
While Marketing has a hand in developing and displaying these graphs, they do generally have a kernel or three of truth, as the devs actually try their best to make them represent what is happening in the real world. After all, why would you want to display a blatant lie in your network operations room, where your technicians might become confused?
The real nuts and bolts of the NOCC is looking at real-time problems that are developing on the network, like when a "region" (a group of servers deployed in a datacenter, though you can have multiple regions in a single datacenter; they're logical groupings, rather than entire racks) goes dark, or when a toplevel DNS server starts to freak out. Based on what problems happen when, you can rapidly react to contact the appropriate people to diagnose and troubleshoot the problem. The problems themselves appear on a proprietary monitoring package that every tech and engineer runs on their desktop, though the problems are logically broken down into groups so that they can be monitored more sanely.
As to the question of why the company is named what it's named, the boilerplate that they give every employee (and have been giving everyone since, I believe, the company's founding) is the Hawaiian word angle. Every new recruit and old hand has heard the same story, and in as much as they care why the company is named what it is, they believe it.
The truly impressive feats of technology aren't Akamai's brute-force deployment of servers throughout the world or visualization models, but rather the thought that goes into where servers are deployed (answer: where they are closest (read: network latency) to the greatest number of end-users, which is invariably in datacenters on or close to the backbones that comprise the Internet), the propagation of real-time metadata on the network (which is distributed via a proprietary daemon that is installed on each and every server that they own), and the DNS tricks that are employed to direct queries to the nearest server that has the content that the end user has just requested. Overall, the engineering is impressive and the deployed infrastructure is daunting to outsiders.
It's a very cool company. I rather wish I was still with 'em.
posted by kniedzw at 5:28 AM on June 8, 2007 [2 favorites]
While Marketing has a hand in developing and displaying these graphs, they do generally have a kernel or three of truth, as the devs actually try their best to make them represent what is happening in the real world. After all, why would you want to display a blatant lie in your network operations room, where your technicians might become confused?
The real nuts and bolts of the NOCC is looking at real-time problems that are developing on the network, like when a "region" (a group of servers deployed in a datacenter, though you can have multiple regions in a single datacenter; they're logical groupings, rather than entire racks) goes dark, or when a toplevel DNS server starts to freak out. Based on what problems happen when, you can rapidly react to contact the appropriate people to diagnose and troubleshoot the problem. The problems themselves appear on a proprietary monitoring package that every tech and engineer runs on their desktop, though the problems are logically broken down into groups so that they can be monitored more sanely.
As to the question of why the company is named what it's named, the boilerplate that they give every employee (and have been giving everyone since, I believe, the company's founding) is the Hawaiian word angle. Every new recruit and old hand has heard the same story, and in as much as they care why the company is named what it is, they believe it.
The truly impressive feats of technology aren't Akamai's brute-force deployment of servers throughout the world or visualization models, but rather the thought that goes into where servers are deployed (answer: where they are closest (read: network latency) to the greatest number of end-users, which is invariably in datacenters on or close to the backbones that comprise the Internet), the propagation of real-time metadata on the network (which is distributed via a proprietary daemon that is installed on each and every server that they own), and the DNS tricks that are employed to direct queries to the nearest server that has the content that the end user has just requested. Overall, the engineering is impressive and the deployed infrastructure is daunting to outsiders.
It's a very cool company. I rather wish I was still with 'em.
posted by kniedzw at 5:28 AM on June 8, 2007 [2 favorites]
Danny Lewin's replacement as CTO was Bill Weihl, at whom's house I stayed last time I was in San Francisco.
So there.
posted by MrMoonPie at 7:16 AM on June 8, 2007
So there.
posted by MrMoonPie at 7:16 AM on June 8, 2007
Curses! You win this time, moonpie!
posted by Dave Faris at 8:47 AM on June 8, 2007
posted by Dave Faris at 8:47 AM on June 8, 2007
Thanks for the background, kniedzw. Out of curiosity, how much extra latency does the overhead for reporting add to the network traffic? And how much data are they backing up?
posted by Civil_Disobedient at 4:47 PM on June 8, 2007
posted by Civil_Disobedient at 4:47 PM on June 8, 2007
hincandenza: I don't work there. Their NOC is on street level, and I walk past their building to get to the subway.
posted by mkb at 2:02 PM on June 10, 2007
posted by mkb at 2:02 PM on June 10, 2007
« Older Japanese historical photographs ca. 1910 | Watch out for meatyfilter Newer »
This thread has been archived and is closed to new comments
posted by Dave Faris at 3:24 PM on June 7, 2007 [1 favorite]