From: Ronald Rojas <ronladred@gmail.com>
To: xen-devel <xen-devel@lists.xen.org>,
Ian Jackson <ian.jackson@eu.citrix.com>,
Wei Liu <wei.liu2@citrix.com>,
George Dunlap <george.dunlap@citrix.com>,
George Dunlap <dunlapg@umich.edu>
Subject: [PATCH RFC 22/59] Use tsc for time rather than rumpkernel clock_gettime()
Date: Wed, 28 Dec 2016 20:14:15 -0500 [thread overview]
Message-ID: <1482974092-15891-22-git-send-email-ronladred@gmail.com> (raw)
In-Reply-To: <1482974092-15891-1-git-send-email-ronladred@gmail.com>
From: George Dunlap <george.dunlap@citrix.com>
The rumpkernel clock_gettime() drifts drastically when the system is
overloaded. Switch to using a TSC timesource instead:
* Use tsc_mode='native' in guest config
* Read the host tsc from /proc/cpuinfo when starting a run
* Pass it as an argument to the worker
* Implement now() with rdtsc
The results appear to continue to be accurate even with heavy
overcommitment.
Signed-off-by: George Dunlap <george.dunlap@citrix.com>
---
benchmark.go | 8 ++++++++
run.go | 45 +++++++++++++++++++++++++++++++++++++++++++++
xenworker.go | 1 +
3 files changed, 54 insertions(+)
diff --git a/benchmark.go b/benchmark.go
index b0b55c7..7ea9aaa 100644
--- a/benchmark.go
+++ b/benchmark.go
@@ -48,6 +48,14 @@ type WorkerParams struct {
Args []string
}
+func (l *WorkerParams) SetkHZ(kHZ uint64) {
+ if l.Args[0] == "kHZ" {
+ l.Args[1] = fmt.Sprintf("%d", kHZ)
+ } else {
+ l.Args = append([]string{"kHZ", fmt.Sprintf("%d", kHZ)}, l.Args...)
+ }
+}
+
type WorkerConfig struct {
Pool string
}
diff --git a/run.go b/run.go
index 2a93405..4222001 100644
--- a/run.go
+++ b/run.go
@@ -23,6 +23,9 @@ import (
"os"
"os/signal"
"time"
+ "regexp"
+ "strconv"
+ "bufio"
)
type WorkerState struct {
@@ -106,9 +109,45 @@ func NewWorkerList(WorkerSets []WorkerSet, workerType int) (wl WorkerList, err e
return
}
+var CpukHZ uint64
+
+func getCpuHz() (err error) {
+ if CpukHZ == 0 {
+ var cpuinfo *os.File
+ cpuinfo, err = os.Open("/proc/cpuinfo")
+ if err != nil {
+ return
+ }
+ re := regexp.MustCompile("^cpu MHz\\s*: ([0-9.]+)$")
+ scanner := bufio.NewScanner(cpuinfo)
+ for scanner.Scan() {
+ s := scanner.Text()
+ m := re.FindStringSubmatch(s)
+ if m != nil {
+ var MHZ float64
+ MHZ, err = strconv.ParseFloat(m[1], 64)
+ if err != nil {
+ return
+ }
+ CpukHZ = uint64(MHZ*1000)
+ break
+ }
+ }
+ if CpukHZ == 0 {
+ err = fmt.Errorf("Couldn't find cpu MHz")
+ return
+ } else {
+ fmt.Println("CpukHZ: ", CpukHZ)
+
+ }
+ }
+ return
+}
+
func (run *BenchmarkRun) Run() (err error) {
for wsi := range run.WorkerSets {
run.WorkerSets[wsi].Config.PropagateFrom(run.WorkerConfig)
+ run.WorkerSets[wsi].Params.SetkHZ(CpukHZ)
}
Workers, err := NewWorkerList(run.WorkerSets, WorkerXen)
@@ -164,6 +203,12 @@ func (run *BenchmarkRun) Run() (err error) {
}
func (plan *BenchmarkPlan) Run() (err error) {
+
+ err = getCpuHz()
+ if err != nil {
+ return
+ }
+
for i := range plan.Runs {
r := &plan.Runs[i];
if ! r.Completed {
diff --git a/xenworker.go b/xenworker.go
index 4077e77..e98c970 100644
--- a/xenworker.go
+++ b/xenworker.go
@@ -95,6 +95,7 @@ func (w *XenWorker) Init(p WorkerParams, g WorkerConfig) (err error) {
fmt.Fprintf(cfg, "memory = 32\n")
fmt.Fprintf(cfg, "vcpus = 1\n")
fmt.Fprintf(cfg, "on_crash = 'destroy'\n")
+ fmt.Fprintf(cfg, "tsc_mode = 'native'\n")
if g.Pool != "" {
fmt.Fprintf(cfg, "pool = '%s'\n", g.Pool)
--
2.7.4
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
next prev parent reply other threads:[~2016-12-29 1:14 UTC|newest]
Thread overview: 67+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-12-29 1:13 [PATCH RFC 01/59] Initial controller framework Ronald Rojas
2016-12-29 1:13 ` [PATCH RFC 02/59] controller: Revamp communication structure Ronald Rojas
2016-12-29 1:13 ` [PATCH RFC 03/59] controller: Initial attempt to generalize process / vm creation Ronald Rojas
2016-12-29 1:13 ` [PATCH RFC 04/59] Controller: Move process worker into its own file Ronald Rojas
2016-12-29 1:13 ` [PATCH RFC 05/59] controller: Add WorkerParams argument to Init in Worker interface Ronald Rojas
2016-12-29 1:13 ` [PATCH RFC 06/59] Reorganize to enable "Dist" directory Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 07/59] controller: Introduce basic Xen functionality Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 08/59] controller: Exit after second SIGINT Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 09/59] controller: Refactor creation and stopping of workers into WorkerList methods Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 10/59] controller: First cut at BenchmarkParams Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 11/59] Refactor to move towards benchmark "plans" and data analysis Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 12/59] Basic 'report' functionality Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 13/59] Add GPL headers / COPYING file (v2 only) Ronald Rojas
2016-12-29 10:51 ` Wei Liu
2016-12-29 1:14 ` [PATCH RFC 14/59] benchmark: Store data in terms of worker sets and worker ids Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 15/59] controller: Move "running" code to a separate file Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 16/59] controller: Rename an element in BenchmarkRun to be more accurate Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 17/59] controller: Collect and display statistics on WorkerSets Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 18/59] controller: Add cpupool global config Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 19/59] Add basic libxl framework, get domain cpu_time Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 20/59] xenworker: Use libxl_domain_unpause rather than forking xl Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 21/59] Report utilization statistics Ronald Rojas
2016-12-29 1:14 ` Ronald Rojas [this message]
2016-12-29 1:14 ` [PATCH RFC 23/59] run: Don't collect results reported after command to stop guests is issued Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 24/59] report: Lots of changes Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 25/59] main: Change default workload to something a bit more extreme Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 26/59] Use kops rather than mops Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 27/59] report: Allow report verbosity to be specified Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 28/59] controller: Handle worker early death Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 29/59] report: Add basic html report Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 30/59] htmlreport: Include utilization scatterplots Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 31/59] Make a binary that can run reports on a system without libxl Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 32/59] controller: Allow specification of an input file Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 33/59] controller: Add verbosity argument and update README with new instructions Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 34/59] controller: Make a useful config file Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 35/59] libxl: Add ListCpupool Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 36/59] controller: Make 'dummy' at the level of 'run' rather than xenworker Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 37/59] libxl.go: Provide a single global context by default Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 38/59] controller: Allow multiple schedulers in the same benchmark file Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 39/59] libxl.go: Put common link flags in libxl.go Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 40/59] controller: Add / update GPL text Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 41/59] libxl.go: Link statically rather than dynamically Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 42/59] plan: Allow "templating" from other runs Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 43/59] libxl: Add bitmap support Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 44/59] libxl: Implement CpupoolCreate Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 45/59] libxl: Implement Destroy, Add/Remove operations Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 46/59] libxl: Reorganize bitmapGotoC Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 47/59] libxl: Reorganize code Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 48/59] libxl: Add Ctx.CheckOpen Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 49/59] libxl: Implement libxl_cpupool_info and Scheduler.FromString() Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 50/59] libxl: Fix Bitmap.Max(), make Test() / Clear() more robust Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 51/59] controller: Make and/or modify cpupools when possible Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 52/59] libxl: Implement Bitmap.String() Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 53/59] controller: Add WorkerConfig.SoftAffinity Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 54/59] controller/run: Add RunConfig.NumaDisable Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 55/59] plan: Make the matrix generation more programmatic Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 56/59] controller/plan: Add NumaDisable to SimpleMatrix Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 57/59] tools/blktap2: remove unused inclusion of sys/sysctl.l Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 58/59] remove irrelevant files from old repository Ronald Rojas
2016-12-29 1:14 ` [PATCH RFC 59/59] tools/xenlight: Create interface for xenlight info Ronald Rojas
2016-12-29 10:34 ` George Dunlap
2016-12-29 10:52 ` Wei Liu
2016-12-29 13:49 ` Ronald Rojas
2016-12-29 13:45 ` George Dunlap
2016-12-29 20:20 ` George Dunlap
2017-01-03 17:45 ` Ronald Rojas
2017-01-04 16:44 ` George Dunlap
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1482974092-15891-22-git-send-email-ronladred@gmail.com \
--to=ronladred@gmail.com \
--cc=dunlapg@umich.edu \
--cc=george.dunlap@citrix.com \
--cc=ian.jackson@eu.citrix.com \
--cc=wei.liu2@citrix.com \
--cc=xen-devel@lists.xen.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).