Wow - great question, I thought. However, I was disappointed to see his results:
26% - Our own internal Internet bandwidth or connectivity
26% - Actual application/system design of vendor's application (JRS - huh??)
21% - Overall Internet traffic
15% - Cloud vendor's Internet connectivity
10% - Don't know
2% - Other
========================================
I found this survey to be a warm, juicy, smoking pile of crap.
... Deep breath ... ... Sip of coffee ...
Computing infrastructure is generally considered a 3-legged stool: 1. Compute power (cpu/memory) 2. Network and 3. Storage. On top of this we lay down our software architectures (n-tier, interpreted/compiled, etc.) If you're going to look for a performance problem, this is a decent way to slice it up. What you don't do is offer participants 3 answers that all lead to "NETWORK". Although, I did enjoy my course in college on, "Misusing Statistics to Your Advantage". Ah - those were the days.
Now, I don't have any fancy pie chart - or an actual survey... but I do talk with LOTS of chief architects on their cloud performance. Here's what I hear:
1. I'm happy with the performance
-- (yes, we did some minor rearchitecting to take advantage of cloud properties)
2. We're not going to the cloud for performance reasons.
-- (These guys typically have high performance SAN's and feel like they'll melt down the DB/messaging tier; and yes, cloud providers are working on this)
3. The application didn't perform 1:1 relative to our on-premise due to the cloud virtualization
-- We solved this by adding another node to the tier.
4. Our application is a web of modules and we only moved some of them to the cloud. Now we see latency issues between on-premise and cloud deployed modules.
- First, if it's a 'real' company, they already have a dedicated line to the cloud provider. Second, they're probably reviewing the distributed computing chattiness and trying to figure out if they can change the message granularity, caching, etc.
Mike continues to discuss WAN optimization as it relates to cloud (JRS - ???) and other jibber. People - there are real issues related to cloud performance. Perhaps it's time to pull out your "Introduction to Computer Science" book and throw away your copy of InformationWeek.