Skip to content

Commit

Permalink
DISKSPD 2.2 (#209)
Browse files Browse the repository at this point in the history 
* add sorted data to Historgram (#197)

Co-authored-by: Feng Li <fengli@microsoft.com>

* multiple updates from msft internal
2.1.2 prerelease

flush stdout to force XML/text results through prior to process exit
fix issues identified by vs2019 build loop - 64->32 downcast, minor printf issues, _snprintf -> _snprintf_s
timestamp prefixes for all verbose output
dump specific time intervals for warmup/measured/cooldown to validate v. expected @ actual time of use/measurement
only emit warmup/cooldown verbose output notes if there is nonzero warmup/cooldown specified in the profile
minor cleanup of load thread verbose output so that everything has a consistent thread N: prefix
histogram processing speedups, about 60% for XML (all percentiles)
implicitly seal/sort histogram on first read operation, reset on subsequent inserts; save repeated re-sorting
save histogram percentile iterator so that asecending percentile queries do not restart from lowest sample
min/max in terms of begin/rbegin iterator
show latency histogram bucket counts in XML results; potential work to moderate histogram size
unit test coverage

* DISKSPD 2.2

---------

Co-authored-by: FengLiMS <128095530+FengLiMS@users.noreply.github.com>
Co-authored-by: Feng Li <fengli@microsoft.com>
Co-authored-by: Dan Lovinger <danlo@ntdev.microsoft.com>
  • Loading branch information
@dl2n @FengLiMS
4 people authored Jun 13, 2024
1 parent 206bae5 commit 2ac91a4
Show file tree
Hide file tree
Showing 41 changed files with 13,862 additions and 12,590 deletions.
4,029 changes: 2,070 additions & 1,959 deletions CmdLineParser/CmdLineParser.cpp
View file

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion CmdRequestCreator/diskspd.rc
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#include <windows.h>
#include "Common.h"
#include "Version.h"

DISKSPD.XSD HTML "..\\XmlProfileParser\\diskspd.xsd"

Expand Down
69 changes: 49 additions & 20 deletions Common/Common.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ TRACELOGGING_DEFINE_PROVIDER(g_hEtwProvider,
(0xca13db84, 0xd0a9, 0x5145, 0xfc, 0xa4, 0x46, 0x8d, 0xa9, 0x2f, 0xdc, 0x2d));

SystemInformation g_SystemInformation;
ULONG g_ExperimentFlags;

UINT64 PerfTimer::GetTime()
{
Expand Down Expand Up @@ -99,12 +100,12 @@ UINT64 PerfTimer::SecondsToPerfTime(const double seconds)
Random::Random(UINT64 ulSeed)
{
UINT32 i;

_ulState[0] = 0xf1ea5eed;
_ulState[1] = ulSeed;
_ulState[2] = ulSeed;
_ulState[3] = ulSeed;

for (i = 0; i < 20; i++) {
Rand64();
}
Expand Down Expand Up @@ -138,10 +139,10 @@ void Random::RandBuffer(BYTE *pBuffer, UINT32 ulLength, bool fPseudoRandomOkay)
pBuffer += Remaining * 8;

if (fPseudoRandomOkay) {

//
// Generate 5 random numbers and then mix them to produce
// 16 random (but correlated) numbers. We want to do 16
// 16 random (but correlated) numbers. We want to do 16
// numbers at a time for optimal cache line alignment.
// Only do this if the caller is okay with numbers that
// aren't independent. A detailed analysis of the data
Expand All @@ -150,7 +151,7 @@ void Random::RandBuffer(BYTE *pBuffer, UINT32 ulLength, bool fPseudoRandomOkay)
// instance it's unlikely compression algorithms will be
// able to detect this and utilize it).
//

while (Remaining > 16) {
r1 = Rand64();
r2 = Rand64();
Expand Down Expand Up @@ -195,7 +196,7 @@ void Random::RandBuffer(BYTE *pBuffer, UINT32 ulLength, bool fPseudoRandomOkay)
//
// Fill in the tail of the buffer
//

while (Remaining >= 4) {
r1 = Rand64();
r2 = Rand64();
Expand Down Expand Up @@ -241,7 +242,7 @@ string ThreadTarget::GetXml(UINT32 indent) const
{
char buffer[4096];
string sXml;

AddXmlInc(sXml, "<ThreadTarget>\n");

sprintf_s(buffer, _countof(buffer), "<Thread>%u</Thread>\n", _ulThread);
Expand All @@ -262,7 +263,7 @@ string Target::GetXml(UINT32 indent) const
{
char buffer[4096];
string sXml;

AddXmlInc(sXml, "<Target>\n");
AddXml(sXml, "<Path>" + _sPath + "</Path>\n");

Expand Down Expand Up @@ -295,7 +296,7 @@ string Target::GetXml(UINT32 indent) const
AddXml(sXml, "<WriteThrough>true</WriteThrough>\n");
break;
}

// MemoryMappedIoMode::Off is implied default
switch (_memoryMappedIoMode)
{
Expand Down Expand Up @@ -416,7 +417,7 @@ string Target::GetXml(UINT32 indent) const
{
sprintf_s(buffer, _countof(buffer), "<StrideSize>%I64u</StrideSize>\n", GetBlockAlignmentInBytes());
AddXml(sXml, buffer);

AddXml(sXml, _fInterlockedSequential ?
"<InterlockedSequential>true</InterlockedSequential>\n" :
"<InterlockedSequential>false</InterlockedSequential>\n");
Expand Down Expand Up @@ -672,6 +673,13 @@ string Profile::GetXml(UINT32 indent) const
sprintf_s(buffer, _countof(buffer), "<Progress>%u</Progress>\n", _dwProgress);
AddXml(sXml, buffer);

if (g_ExperimentFlags)
{
// only output if on so that downlevel doesn't get (and fail: not in downlevel xsd) unless actually specified
sprintf_s(buffer, _countof(buffer), "<ExperimentFlags>%u</ExperimentFlags>\n", g_ExperimentFlags);
AddXml(sXml, buffer);
}

if (_resultsFormat == ResultsFormat::Text)
{
AddXml(sXml, "<ResultFormat>text</ResultFormat>\n");
Expand All @@ -686,6 +694,12 @@ string Profile::GetXml(UINT32 indent) const
}

AddXml(sXml, _fVerbose ? "<Verbose>true</Verbose>\n" : "<Verbose>false</Verbose>\n");
if (_fVerboseStats)
{
// only output if on so that downlevel doesn't get (and fail: not in downlevel xsd) unless actually specified
AddXml(sXml, "<VerboseStats>true</VerboseStats>\n");
}

if (_precreateFiles == PrecreateFiles::UseMaxSize)
{
AddXml(sXml, "<PrecreateFiles>UseMaxSize</PrecreateFiles>\n");
Expand Down Expand Up @@ -763,7 +777,7 @@ bool Profile::Validate(bool fSingleSpec, SystemInformation *pSystem) const
}
if (fOk && !pSystem->processorTopology._vProcessorGroupInformation[Affinity.wGroup].IsProcessorValid(Affinity.bProc))
{
fprintf(stderr, "ERROR: affinity assignment to group %u core %u not possible; group only has %u cores\n",
fprintf(stderr, "ERROR: affinity assignment to group %u cpu %u not possible; group has a max of %u cpus\n",
Affinity.wGroup,
Affinity.bProc,
pSystem->processorTopology._vProcessorGroupInformation[Affinity.wGroup]._maximumProcessorCount);
Expand All @@ -773,22 +787,27 @@ bool Profile::Validate(bool fSingleSpec, SystemInformation *pSystem) const

if (fOk && !pSystem->processorTopology._vProcessorGroupInformation[Affinity.wGroup].IsProcessorActive(Affinity.bProc))
{
fprintf(stderr, "ERROR: affinity assignment to group %u core %u not possible; core is not active (current mask 0x%Ix)\n",
fprintf(stderr, "ERROR: affinity assignment to group %u cpu %u not possible; cpu is not active (current mask 0x%p)\n",
Affinity.wGroup,
Affinity.bProc,
pSystem->processorTopology._vProcessorGroupInformation[Affinity.wGroup]._activeProcessorMask);
(void *) pSystem->processorTopology._vProcessorGroupInformation[Affinity.wGroup]._activeProcessorMask);

fOk = false;
}
}
}

// ISSUE: many of the following validation errors are stated in cmdline terms, which is not helpful for XML

if (timeSpan.GetDisableAffinity() && timeSpan.GetAffinityAssignments().size() > 0)
{
fprintf(stderr, "ERROR: -n and -a parameters cannot be used together\n");
fOk = false;
}

// ISSUE: with XML and the following the target specification validation it would be useful to say what
// target they're for

for (const auto& target : timeSpan.GetTargets())
{
const bool targetHasMultipleThreads = (timeSpan.GetThreadCount() > 1) || (target.GetThreadsPerFile() > 1);
Expand Down Expand Up @@ -881,7 +900,7 @@ bool Profile::Validate(bool fSingleSpec, SystemInformation *pSystem) const
fprintf(stderr, "ERROR: random distribution ranges (-rd) do not apply to sequential-only IO patterns\n");
fOk = false;
}

if (target.GetUseParallelAsyncIO() && target.GetRequestCount() == 1)
{
fprintf(stderr, "WARNING: -p does not have effect unless outstanding I/O count (-o) is > 1\n");
Expand Down Expand Up @@ -958,7 +977,7 @@ bool Profile::Validate(bool fSingleSpec, SystemInformation *pSystem) const
// Note that absolute range needs no additional validation - known nonzero/large enough for IO
if (target.GetDistributionType() == DistributionType::Percent)
{
if (targetAcc + r._dst.second > 100)
if (targetAcc + r._dst.second > 100)
{
fprintf(stderr, "ERROR: invalid random distribution Target%% %I64u: can be at most %I64u - total must be <= 100%%\n", r._dst.second, 100 - targetAcc);
fOk = false;
Expand Down Expand Up @@ -1036,10 +1055,20 @@ bool Profile::Validate(bool fSingleSpec, SystemInformation *pSystem) const
}
}

// in the cases where there is only a single configuration specified for each target (e.g., cmdline),
// currently there are no validations specific to individual targets (e.g., pre-existing files)
// so we can stop validation now. this allows us to only warn/error once, as opposed to repeating
// it for each target.
// Note that this error is only possible with -f or XML. The -Bbase:length form is immune.
if (target.GetMaxFileSize() && target.GetMaxFileSize() <= target.GetBaseFileOffsetInBytes())
{
fprintf(stderr, "ERROR: maximum (-f) target offset must be greater than base (-B)\n");
fOk = false;
}

// If we know there is only a single target specification (the parameters which apply to targets) shared
// across the one or more targets, we can stop. In practical terms this is the command line case - for
// XML we don't know, and do need to keep going. This early exit lets us avoid repeating the same sets
// of error messages per each target we would otherwise loop over.
//
// If we ever did target property validation (say, v. its size) we'd want to divide out the validations
// into parameter-only v. parameter/property cases for similar reasons.
if (fSingleSpec)
{
break;
Expand Down Expand Up @@ -1113,7 +1142,7 @@ BYTE* ThreadParameters::GetReadBuffer(size_t iTarget, size_t iRequest)
BYTE* ThreadParameters::GetWriteBuffer(size_t iTarget, size_t iRequest)
{
BYTE *pBuffer = nullptr;

Target& target(vTargets[iTarget]);
size_t cb = static_cast<size_t>(target.GetRandomDataWriteBufferSize());
if (cb == 0)
Expand Down
Loading

0 comments on commit 2ac91a4

Please sign in to comment.