Are Your Cloud Computing Platforms Obsolete?
April 25, 2012
Every cloud computing platform that is being sold today will be obsolete prematurely unless they can retrofit them with a single-source timing device. If cloud computing is going to be as pervasive tomorrow as some sales executives have hyped them, the need for a more sophisticated platform has to be fulfilled today.
It’s not enough to tweak some components or put some functions like I/O on the chip. That will definitely help performance, but we are not looking at just shaving off some latency when it comes to financial and other mission-critical applications. Clocking needs to be totally synchronous and that means getting it from one source…
The importance of cloud computing is being "amped up" because manufacturers have to have some battle cry to boost sales of servers as well as next-generation chips. Would you buy this year’s car model if they only had eight-track tape systems in them? Half of you probably don’t even know what an eight-track tape is. The short answer is "No." You would wait until they made the car with the proper instrumentation.
The same logic should be applied to servers. If I am to build a sophisticated server farm that provides applications to a diverse number of subscribers, I need to make sure I have the latest and greatest capabilities. One of those capabilities is the ability to have a single source for accurate timing across the cloud or for that matter, several clouds.
Timing Is Everything
In an earlier article, I discussed the importance of Timing-as-a-Service"©
Timing from a single source is one of the missing elements that cloud computing needs to incorporate into its fabric in order to provide a solid platform for enterprise mission-critical applications.
No vendor seems to have a server that has a single-source timing capability where you could bridge across to different clouds together. Some salespeople don’t even understand the concept, so how can they formulate the solution?
You cannot develop certain types of mission-critical applications, especially financial ones, where they depend on a very accurate timestamp, unless you have the ability to synchronize various platforms together. To be very specific, you need to synchronize every cloud to one clock. If you don’t, you have no way of realizing total transaction synchronicity across your application or across the network.
Looking at INTEL’s latest chips and Dell’s offering of their 12th generation servers; they are still lacking what is needed for financial transactions and the HFT market (High Frequency Transactions).
They did add some bells and whistles, but not what I consider the "killer app" for chips. HFT needs what is depicted at the bottom of the Financial Transactions Timing Chart below. The ability to source timing from one source so that everything on the network is synced up to the nanosecond.
Financial Transactions Timing Chart
|
SPEED of TRANSACTIONS (in seconds) |
SECONDS |
TIMING FROM |
OBSERVATION |
|
Hundredth |
1/100 |
MULTIPLE SERVERS |
Where regulators are today |
|
Millisecond |
1/ 1,000 |
MULTIPLE SERVERS |
Latency (30-40 milliseconds) |
|
Microsecond |
1/10,000 |
MULTIPLE SERVERS |
Where traders are today |
|
Nanosecond |
1/1,000,000 |
ONE ATOMIC CLOCK |
Where regulators should be |
Source: James Carlini, 2012
If You Can’t Measure It, You Can’t Manage It
How can you measure transactions and their delivery times with the cloud platforms today? You can’t. And because you can’t, you cannot truly manage any real financial transactions in a multi-member (multi-cloud) community.
When they can get full synchronization of the cloud, then the vendors and manufacturers can make some real money. Until then, beware of platforms that do not provide full cloud transaction synchronicity.