Amazon AWS Review

Accessing apps on AWS via my iPhone is awful. We use it, because it improves the speed for us to access vendors.

What is our primary use case?

To access systems of partner/vendor companies, we maintain an instance to transfer data to our instance, then privately back to us. Basically, a BRAS, B-RAS or BBRAS device.

How has it helped my organization?

It improves the speed for us to access vendors, etc. AWS is extremely slow over the internet. Where we have GigE fiber over dedicated OC48 links, and when ping times to Dallas, TX from San Francisco is 30ms RTT on average, AWS is always 20ms higher. To AWS East, it is 70-110ms RTT, and the data transfer almost seems throttled. So I spun up an instance, made a BRAS image, like how DSL customers access the internet, and set up a peer with AWS to transfer data privately, then publicly from our instance to the AWS IP of our vendor, or partner and it has improved response times dramatically. The average API access latency was 250ms, horribly slow - already authenticated, etc. 

We also use it for our Amazon Seller Account and Amazon Vendor Account, where Amazon's systems run. Amazon recently moved their systems to CloudFront, but AWS DNS is awful slow. So the BRAS helps with the DNS as well.

What is most valuable?

Bidding on instances with dynamic pricing. So, I can do something that is not critical in terms of speed, like a production system, but testing and bid at an uber low price, and I will usually get what I want.

What needs improvement?

The network is way overloaded. Comcast is overloaded. So between the two, it sucks.

I am used to Level (3) or Verizon/ AS701 with fabulous ping times and throughput, where I click something and it works. 

It is the problem with the nomenclature of SDN [software defined networking] as engineers today do not understand networking, TCP/IP, or anything. I was 18 during the .com bust, but I remember accessing tools, as I worked for a Global ISP NTT, which owned Verio, the largest webhost at the time. We had Dual OC-3's to our office, when our office was just a remote NOC, but we had cloud computing before it was nomenclature. We accessed customer data, and had tools to do things quickly, instead of logging into routers, IDS, IPS, switches, and server. If it was a repetitive task, it would be via a browser, and the browser accessed a txn server rather than run cron jobs every 15 minutes. I will say that AWS for API, or Seller Central, is no improvement from what we had (our internal tools we designed to update accounts, change customer network profiles, monitoring, MRTG graphs, etc), when AWS should be blazing. 

Accessing apps on AWS via my iPhone is awful. Apple is behind the times in speed, battery, and even the screen, but it is aesthetically pleasing, so it wins. Android devices by Samsung are superior, but I use iPhone because that is what we use in Silicon Valley, which is on Verizon's LTE Advanced (LTE-X is their coined term) network, and the latency is great, 20-30ms, speeds of 40 Megabits/s, symmetrical are quite common, and sometimes I see 150/150 Megabits/s.  

For how long have I used the solution?

More than five years.

Which other solutions did I evaluate?

When they first moved Amazon Seller Central to AWS CloudFront from AWS, I would see connections to Hong Kong and Singapore. Maybe I was sent there because the USA East was overloaded. I do not know. So, we started using Verisign for recursive DNS, and to host our own domain name(s), and I noticed, it fixed the problem. Every ISP and DNS server, either Unicast like Level(3), Google, NTT was the best performer, Comcast was garbage, while on Comcast network, Verizon was good (FiOS [consumer and SMB], and Enterprise which is what we have aka MCI aka UUNET/ AS701/AS702/AS703), and SprintLink AS1239 was good.  However, we just tested Sprint, and Verizon. Verizon provides backup services for us, actually tertiary, we provide our own secondary.  So I signed us up with Verisign with DDoS protection, made Verizon secondary, and the feed server from us, feeds VeriSign and Verizon. That fixed the AWS CloudFront location issue, which to me, shows how poor AWS DNS is.

We would get responses that are AWS Hong Kong, even when they moved to CloudFront to speed up Seller Central (I complained to corporate via a letter FedEx'ed to Amazon). I asked for a private MPLS link, which we would pay for, and we were told it would be worked on. During peak times, it would lag and time out, it was awful. It still lags, but I route as much as we can via the BRAS setup. 

What other advice do I have?

I have pushed clients towards Microsoft Azure. I have bugged Microsoft to add links to their network in the BNA region, BNA to Atlanta (additional link), BNA to MCI aka Kansas City, BNA to Chicago, BNA to WDC, and BNA to Dallas, TX to improve access for things at BNA. It is not critical. It is just the only facility that is 30ms slower than others. Azure in Chicago, Wyoming, Bellevue, Silicon Valley, Los Angeles, or Texas is very low latency. 

I have also pushed clients to IBM Bluemix, as their partnership with Akamai makes API access is really fast. Azure with Verizon CDN/Terremark is fabulous.

I have to add this. AWS sucks, even though I am a customer.

Disclosure: I am a real user, and this review is based on my own experience and opinions.
Add a Comment
Sign Up with Email