Cloud getting crowded, and that means bottlenecks
As data-transfer shifts increasingly to the cloud, the servers stacked in datacenters handling the data become increasingly crowded. Virtualization means multiple users can share a single server.
This has positive aspects: servers don't sit idle, scalability is less of a concern, and datacenter efficiency improves. But there's a problem with too many users on a single server.
And it's going to get worse.
It's called "the noisy neighbor problem," and here's what happens: disk I/O for one user starts to interfere with the operations of another user on the same server.
It's the I/O. Today's virtualization techniques do a great job of segmenting memory and CPU usage. Flash-drives (SSDs) help speed data bottlenecks. But when one or more virtual machines on the physical host consumes large amounts of disk I/O, performance for other virtual machines degrades.
The symptoms can be subtle. Spotty performance, like service-glitches or scalability problems, can indicate I/O interference from other users on your shared server.
And as more users are put on a physical server through virtualization, the greater the chance of I/O problems.
Here's a worst-case scenario: a database server slows down (it happens). It simply queues the extra queries. But that consumes extra RAM, so the Web servers using that RAM for queries back up with requests. Can you see where this is heading?
Web server threads consume the available RAM waiting for the database, so those hosts resort to swap-files. By jamming the RAM, speed-advantages disappear and disk I/O begins to swamp the server...and all its users.
What to do? Start a conversation with your cloud provider. Mention your concern about "noisy neighbors." Smaller servers with fewer users per server are a plus--ask how many other users are anticipated on your particular server.
And as always, check your SLA.
Cloud service is a competitive business and any provider wanting your custom should be forthcoming with answers.
As cloud computing increasingly becomes commoditized, users need to know answers to specific questions. Due diligence produces queries for responsible vendors.