Skip to main content

Hiding poor performance in processor cycles

Solaris remains the only operating system capable of microstate accounting. Other operating systems like Linux, FreeBSD, and Windows can not accurately report how their kernel spends an application's user mode time and kernel mode time. Although, Linux kernel 3.0 says it supports microstate accounting it only reports approximate time spent as percentages. What that means is applications can hide their poor performance in processor cycles.

Even if your kernel doesn't support microstate accounting you can still investigate your application for poor performance using commands like time, killall, timeout and vmstat.

The command time reports three psuedo time values of a running process: real time, user time, and system time. Only real time is valuable, but it isn't actually real time. Run time using your application's absolute path as its argument:

time <command> 2>&1;

Once your application opens, guide your application to the point you want to check then kill it by sending your application the signal SIGTERM, SIGSTOP or SIGKILL. (You could also start using it as you normally would then kill it when it starts hanging or slowing down.) Run killall to kill all processes owned by your application:

killall -s <signal> <command or pid>;

Run your application again, but this time use your application's real time as the argument for the command timeout along with "vmstat 1". The command timeout runs vmstat every 1 second for the amount of real time. Guide your application to the same point you wanted to check before. The command vmstat will display CPU information every second leading up to your checkpoint including the processing of the kill signal. Here's the command to run:

<command> 2>&1; timeout <real time> vmstat 1 2>&1;

Do you have a suggestion about how to improve this blog? Let's talk about it. Contact me at David.Brenner.Jr@Gmail.com or 720-584-5229.

Comments

Post a Comment

Comments to this blog will be reviewed within 72 hours. No trolling please

Popular posts from this blog

OpenStack+Ceph as Software-Defined Storage

SDS reduces the costs of the management of growing data stores by decoupling storage management from its hardware to allow for centralized management of cheaper, popular commodity hardware. The example SDS ecosystem uses open source software like OpenStack as a front-end interface on top of Ceph as the resource provider of a RADOS cluster of commodity solid-state drives. OpenStack provides user-friendly wrappers for accessing and modifying underlying Ceph storage. OpenStack comes in the form of distributed microservices with RESTful API's: Block (Cinder), File (Manila), Image (Glance), and Object (Swift). Each microservice can scale-out as a cluster of stand-alone services to accommodate the varying demands of high-growth storage. With OpenStack the underlying Ceph storage can address the block storage needs, file storage needs, image storage needs, and object storage needs of datacenters adopting open source as their new norm in an industry trend for high performace and high a...
1 comment
Read more

The meaning of time in reinforcement learning

Reinforcement learning (RL) is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning. Reinforcement learning is concerned with how software agents ought to take actions in an environment in order to maximize the notion of cumulative reward through the process of trial and error. In reinforcement learning an agent starts at an empty state then analyzes the available datasets according to a policy of positive states and negative states. Rather than being explicitly taught as in supervised learning the correct set of actions for performing a task, reinforcement learning uses rewards as signals for positive states and punishments as signals for negative states. The agent obtains the best path to a desirable reward as a cumulation of positive states and negative states. As compared to unsupervised learning, reinforcement learning is different in terms of goals. While the goal in unsupervised learning is to find similarities and differences...
Post a Comment
Read more

Principal Component Analysis

Principal Component Analysis (PCA) is a common technique in statistical analysis, widely used for pattern recognition, data compression, image preprocessing, signal-noise analysis, and high resolution spectrum analysis. Principal Component Analysis transforms a group of activites into a set of unique components, where each component has a numerical degree of distance and relatedness from an agreed on centered component. The first component has the largest possible variance (it accounts for most of the variability in the group). Each succeeding component has the highest variance that is orthogonal to the preceding components. The transformation of the group proceeds linearly from a group with a high degree of dimensionality to a group with a low degree of dimensionality of which the components of the group with a low degree of dimensionality are uncorrelated. Principal Component Analysis is also used in the forecast of a most likely outcome through time-series analysis and regress...
Post a Comment
Read more