Amazon's new CDN

Amazon's new CDN is cheered on by their CTO Werner Vogels. The primary take-away from the post:

Using a global network of edge locations this new service can deliver popular data stored in Amazon S3 to customers around the globe through local access.

Now, Amazon is only currently peering visibly with 4 other networks: Qwest, NTT, Tiscali, and Level3, which all seem to be transit providers.

Now I guess their edge routers don't absolutely need to sit in peered networks, but I'm sure it'll make the host networks happier [otherwise those host networks will be bearing the cost of reverse-proxying the traffic through their own transit connections].

The PeeringDB page for Amazon isn't that much more helpful - no public switching presence noted, but there is some promise shown in the following locations:

• Equinix Ashburn
• PAIX Palo Alto [incl. PacketExchange I guess]
• The Westin Building [Yay SIX! And PAIX Seattle too, I suppose]

Another development to watch closely. Amazon: where is your non-USA PoP presence? LINX and AMS-IX beckon...

Gro'in' pains

Enough 80's references, let's skip to the meat of this post (so to speak). This is from a recent traceroute; DataHop might want to rename one of their nodes to accomodate some expansion in their traffic:

11 196.32.209.25 (196.32.209.25) 165.620 ms 160.809 ms
12 ge2-20-0-cr0.tch.uk.as6908.net (78.41.155.161) 236.578 ms 198.183 ms
13 te-4-2.r00.londen05.uk.bb.gin.ntt.net (83.231.146.69) 151.840 ms 153.098 ms

Unscheduled scheduled outages

A recent posting to the TENET ren-news list:

All Neotel MPLS GEN3 sites experienced a number of service failures
early yesterday morning. It appears that these were caused by Neotel
maintenance, of which TENET was not informed in advance.

even given

The Service Level Agreement provides that: "...The Providers will give TENET four days' advance notice of planned maintenance."

Let's hope Neotel doesn't become Neotelkom [the new Telkom?] :)

150 GB DSL caps?

From the latest DSL Prime:

My follow-up article will be unreliable sources, which I'm factchecking more carefully. For now, a few types to distrust:

• The FCC or anyone else who takes seriously a 200K definition of broadband.
• Anyone who defends a bandwidth cap below about 150 gigabytes per month, except where bandwidth costs are unusually high or the service selling for under about $20. (India, South Africa, possibly the U.K.). Ask mo [sic] for the numbers if you don't get it.

I guess for South Africa, he's specifically talking about high bandwidth costs, 'cos I don't know of any ISPs selling DSL for $20/R160 here.

GEN3 failure

And so soon after the start of the service?

Hmmm, what does the SLA say about this? 99% uptime implies 1% downtime.

365 days x 24 hours x 1% = 87.60 hours down per year.

So they have loads of downtime left - bah humbug.

A quick estimation from the ML Sultan campus of DUT:


The interruption starts around 19h30 on the 5th of March, and lasts until about 09h30 on the 6th of March. That's 14 hours total. Ouch.

So we can look forward to another 73.6 hours of downtime. That's another 3 days left!

GEN3 draft SLA

GEN3's draft SLA is available (and claims to have been posted on 27th February).
Interesting to see specific figures. I'm intrigued to see that 260 ms RTT 'average'.

5. Availability

The Providers commit to the following levels of service element availability:

Services provided by Neotel:99%

Services provided by Internet Solutions:

International services: 99.8%

National services: 99.95%

6. Response times (“RTT”)

The Providers commit to the following network RTT averages:

Services provided by Neotel: 150 ms

Services provided by Internet Solutions:

International services: 260 ms

National services: 35 ms (services provided out of Rosebank)

50 ms (services provided out of Cape Town)

7. Packet loss

The providers commit to confining packet loss to the following levels:

Services provided by Neotel: 1%

Services provided by Internet Solutions:

International services: 1.5%

National services: 2%

Developments on GEN3 and SANREN

According to the latest Rhodes IT minutes, they expect to increase their bandwidth radically:
4.5 Mbps of international bandwidth x a 35 fold increase in ... Internet bandwidth yields 155 Mbps - an OC3. In 2008! At least South Africa isn't too far behind the times ;)

Does this mean UCT will be getting a similar capacity, or something a little faster? Since we'll be near one of the cores of SANREN, hopefully things will be a bit better.

Also, ITWeb thinks SANREN will be a 10 Gbps network (presumably per link, not in aggregate). Note also the proximity to JINX - that should offload some of the transit requirements to direct peering. Also, the projected operational date is end 2008! (a bit earlier than previously thought, and also overlapping with GEN3's mandate with TENET).

Also also, the Infraco weirdness with cabling up the west coast to Europe and South America should matter a little less with the redundant path on SEACOM that TENET has purchased the rights to use.

So, perhaps academic and high speed networking in South Africa has a bright future after all?

AMS-IX breaks 300 Gbps

AMS-IX finally breaks the 300 Gbps mark.

And they'd been stuck on 287 Gbps for so long I thought there was an infrastructure-related limit.

Considering they announced reaching 250 Gbps in February, this implies a compound growth rate of about 3.5%-4% per month.

[And from a presentation POV, interesting to note that the new baseline is not 0, but 100 Gbps! This is more than any exchange not in the top 5!)

I call shotgun!

Internet service providers in the U.S. experienced a service slowdown Monday after fiber-optic cables near Cleveland were apparently sabotaged by gunfire. [Emphasis mine -Al]

The report then goes on to name three different ISPs/NSPs that were affected:

• Level3 (who owns the cable)
• TeliaSonera, and
• Cogent

Don't these people have redundant circuits? Does Level3 have cheap non-protected circuits on offer? Or are some networking companies smart enough to build their own infrastructure on Level3's dark fibre and yet simultaneously dumb enough to not protect that same infrastructure? Wow.

Alternatively, it's a great excuse for more Cogent de-peering *ahem*...
(thanks PCWorld / Yahoo!)