performance - Windows CPU Scheduler - very high kernel time -

-

we trying understand how windows cpu scheduler works in order optimize our applications achieve maximum possible infrastructure/real work ratio. there's things in xperf don't understand , ask community shed light on what's happening. started investigate these issues when got reports servers "slow" or "unresponsive".

background information

we have windows 2012 r2 server runs our middleware infrastructure following specs.

we found concerning 30% of cpu getting wasted on kernel, started dig deeper.

task manager view

the server above runs "host" ~500 processes (as windows services), each of these "host" processes has inner while loop ~250 ms delay (yuck!), , each of "host" processes may have ~1..2 "child" processes executing actual work.

while having infinite loop 250 ms delay between iterations, actual useful work "host" application execute may appear every 10..15 seconds. there's lot of cycles wasted unnecessary looping.

we aware design of "host" application sub-optimal, least, applied our scenario. application getting changed event-based model not require loop , therefore expect significant reduction of "kernel" time in cpu utilization graph.

however, while investigating problem, we've done xperf analysis raised several general questions windows cpu scheduler unable find clear/concise explanation.

what don't understand

below screenshot 1 of xperf sessions.

xperf session

you can see "cpu usage (precise)"

  • there's 15 ms time slices, of majority under-utilized. utilization of slices ~35-40%. assume in turn means cpu gets utilized ~35-40% of time, yet system's performance (let's observable through casual tinkering around system) really sluggish.

  • with have "mysterious" 30% kernel time cost, judged task manager cpu utilization graph.

  • some cpu's utilized whole 15 ms slice , beyond.

questions

as far windows cpu scheduling on multiprocessor systems concerned:

  • what causes 30% kernel cost? context switching? else? consideration should made when applications written reduce cost? or - achieve perfect utilization minimal infrastructure cost (on multiprocessor systems, number of processes higher number of cores)
    • what these 15 ms slices?
    • why cpu utilization has gaps in these slices?

to diag cpu usage issues, should use event tracing windows (etw) capture cpu sampling data (not precise, useful detect hangs).

to capture data, install windows performance toolkit, part of windows sdk.

enter image description here

now run wprui.exe, select first level, under resource select cpu usage , click on start.

enter image description here

now capture 1 minute of cpu usage. after 1 minute click on save.

now analyze generated etl file windows performance analyzer drag & drop cpu usage (sampled) graph analysis pane , order colums see in picture:

enter image description here

inside wpa, load debug symbols , expand stack of system process. in demo, cpu usage comes nvidia driver.


Comments

Post a Comment

Popular posts from this blog

java - nested exception is org.hibernate.exception.SQLGrammarException: could not extract ResultSet Hibernate+SpringMVC -

-
i trying fetch records database , having above error this classs appropriate getters , setters public class user { private int id; private string username; private string password; private string email; i have create database , inserted data below configuration <bean id="datasource" class="org.apache.commons.dbcp2.basicdatasource" destroy-method="close"> <property name="driverclassname" value="com.mysql.jdbc.driver"/> <property name="url" value="jdbc:mysql://localhost:3306/usersdb"/> <property name="username" value="root"/> <property name="password" value=""/> </bean> this complete stacktrace severe: servlet.service() servlet [appservlet] in context path [/spring] threw exception [request processing failed; nested exception org.hibernate.exception.sqlgrammarexception: not ...
Read more

sql - Postgresql tables exists, but getting "relation does not exist" when querying -

-
Image
i have postgresql db number of tables. if query: select column_name information_schema.columns table_name="my_table"; i list of columns returned properly. however, when query: select * "my_table"; i error: (programmingerror) relation "my_table" not exist 'select *\n "my_table"\n' {} any thoughts on why can columns, can't query table? goal able query table. you have include schema if isnt public one select * <schema>."my_table" or can change default schema show search_path; set search_path my_schema; check table schema here select * information_schema.columns for example if table on default schema public both works ok select * parroquias_region select * public.parroquias_region but sectors need specify schema select * map_update.sectores_point
Read more

asp.net mvc - breakpoint on javascript in CSHTML? -

-
i have .cshtml file has javascript tag. within script tag have few lines of javascript. first line of javascript refers property in @viewbag. other javascript lines plain script without mvc related references. observation: can place breakpoints (visual studio 2015) on each of these javascript lines. however, breakpoints has white dot in center of breakpoint symbol (red circle) except line @viewbag reference (this particular line not have white dot in center). it seems breakpoint without white dot hit during run-time. question: can explain going on here? white dot mean? why plain red breakpoint line line hitting breakpoint? the visual studio debugger expecting debug actual server-side code within .cshtml file opposed client-side javascript within it. iirc, visual studio allow debug javascript code expected in internet explorer (and possibly edge), other browsers want use either third-party tool or developer tools (f12) within browser. an easy approach take advant...
Read more