Cloud computing, removing the traffic jams.


I’ve just been reading about the impact of under powering storage area networks on virtualization. I’m going to extend this to cloud computing and look at the real traffic jam points in a cloud.

The news is full of high profile announcements of “cloud processors” (another fine example of “cloudwashing”, attach the term “cloud” to anything). Intel has just come out with an 8 core CPU, AMD is following shortly with 12 cores, Intel and others have been mucking around with 48 or more cores. Where very few applications have ever managed to use a significant number of CPU cores efficiently, the hypervisor is finally the best option for putting lots of cores to work. Unfortunately, CPU is generally not the bottleneck in cloud computing, making the high density multicore CPUs expensive and generally underworked.

Looking at my own cloud today (AMD Opteron based), the utilization of memory is running on average 20 to 30 times the utilization of the CPUs. This is the first bottleneck, memory. You are far better off in most cases with modestly sized CPUs and pack just as much memory into the servers as possible. It’s certainly true that having more CPU cores can allow you to support more virtual machines, but if you can’t provision the memory to support them (and I think this is going to get harder and harder with the growth of CPUs), the extra cores are wasted.

The back end storage system of the cloud computing system is the next largest jam. In general, far more traffic is going to be exchanged over the SAN than the public facing interfaces of the cloud. Increasing the speed of the SAN (note that I said speed, not capacity) has the single biggest effect on the user experience of the cloud computing service.

The load on the SAN also ties directly to the memory available as well. Starving virtual machines for memory and making them swap to disk across the SAN will produce absolutely horrendous results with consequences that will cascade across the cloud affecting everything else running.

So, unless you’re running massive scientific calculation jobs or doing high end animation rendering, take a pass on the gold plated “cloud CPUs”, throw as much memory as possible at your cloud servers, and equip your SAN for maximum performance and keep your cloud from raining on you.

Don’t sit around in a traffic jam waiting for your cloud to pay attention to you, email or call me or visit the SwiftWater Telecom web site for cloud computing services!

Vern

swiftwater telecom rcs cloud computing logo

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s