C# something changed
- Date
- 26 dez 2024 - update to add new thread-pool features
- 11 dez 2024 - setup of the initial document
INDEX
I have to admit, I like C#
Many things that are hard to do in another language are easy to do in C#. Integrating MONO into the NHI1-build-environment was child's play and otherwise I can only report positive things.
- microsoft's documentation is exemplary and has left no question unanswered.
- MONO is mature, has a lot of features and is supported by microsoft.
- MONO integrate very well into the UNIX os and commandline.
Now I unfortunately have to report that MONO has been sold to microsoft, which would be completely legitimate if MONO still existed.
- OpenSUSE has already not integrated the latest MONO packages.
- microsoft itself has announced that .NET is now the actual distribution and MONO is hardly maintained anymore.
The biggest difference between MONO and .NET is the build environment, which is completely different, which logically also has a consequence with the integration into the NHI1-build-environment.
There are two basic types of build environments :
- An active environment that aims to offer a holistic solution that does not aim to integrate into an external solution and thus maximizes the integration wall :
- A passive environment that aims to integrate into an external solution and thus minimizes the integration wall :
Since, as described :
THREADS
A thread is always a problem for so-called "high-level languages", i.e. everything outside of C and C++.
- C# with MONO was an exception in this regard, especially under the version 6.8.0 supported by OpenSUSE.
Now MONO has been sold to microsoft, which initially favors a switch to their C# implementation .NET.
- MONO with the current version 6.13.0 is still supported, but rather hesitantly
To make a long story short:
- The thread support for OpenSUSE mono-6.8.0 is good and met my expectations.
- The thread support for the current mono-6.13.0 is bad and falls behind in the mass test.
summary:
- Under LINUX with mono-6.8.0 the thread is good, but it is no longer supported.
- Under LINUX with mono-6-13.0 the thread support is unusable.
- I can't say whether it will be better with .NET because I haven't tested it.
THREAD POOL
- Example: CS thread pool using
MqSysServerThreadFinterface -
namespace csmqmsgque {public static partial class MqMsgque {static unsafe private void MqSetupTmpl() {Mq.MqSetup();Mq.MqCsInit(MqSysServerThreadCB,MkSysExitCB);}MkErrorEIntPtr mkrt, IntPtr ctx, IntPtr argP, String name, int state, MkIdS *idP) {try {var thread = Task.Factory.StartNew( () => {Mq.MqSysServerThreadMain(argP);});(*idP).type = MkIdSE.THREAD;if ((*idP).ioIsPipe) {(*idP).val = (IntPtr)GCHandle.Alloc(thread,GCHandleType.Weak);} else {(*idP).val = (IntPtr) 0x1;}} catch (Exception ex) {return MkErrorC.FORMAT().Catch(ex).GetCode();}return MkErrorE.OK;MkIdSE}private static void MkSysExitCB ( int isThread, int num ) {if (isThread != 0) {// do nothing, end of MqSysServerThreadMain finish thread} else {Environment.Exit(num);}// FINISH}
- Example: CS context setup
- MqExitF and MqCtxTypeS::MqProcessExitCB
- MqExitF and MqCtxTypeS::MqThreadExitCB
- MqWaitForPipeF and MqCtxTypeS::MqWaitForPipeCB
namespace csmqmsgque {// USER DEFINED CONFIGpublic partial class MqContextC{// START class configstatic MqContextC() {// START type configIntPtr ident_cstr = Marshal.StringToHGlobalAnsi("csmqmsgque_MqContextC");unsafe {MqContextCTT = Mq.MqCsContextTypeCreate(mkrt, ident_cstr,fDefaultLinkCreate, null, fDefaultLinkCreate, null,ProcessExit, ThreadExit, WaitForPipe);}Marshal.FreeHGlobal(ident_cstr);// END type config#define MqContextCTT// END class config}private static void ProcessExit (IntPtr mkrt, int num) {Environment.Exit (num);}private static void ThreadExit (IntPtr mkrt, int num) {}switch ((*idP).type) {case MkIdSE.PROCESS: {// attention: theKernel/c/sys_mk.c -> SysWaitForProcess -> MK_ID_PROCESSvar p = Process.GetProcessById((*idP).val.ToInt32());p.WaitForExit(200);break;}case MkIdSE.THREAD: {var hdl = GCHandle.FromIntPtr((*idP).val);var task = (Task) hdl.Target;if (task != null) {task.Wait();}hdl.Free();break;}case MkIdSE.UNUSED: {break;}}}
REFERENCES
In Programming-Language-Micro-Kernel (PLMK), the connection to a language such as C# is regulated by a reference.
- In C#, the pointer to the PLMK object is stored under
MkObjectC::phdl. - In Programming-Language-Micro-Kernel (PLMK), the pointer to the C# object is stored under
MkObjectS::self.
The pointer of a C# object is stored as week or strong reference :
- weak
- A weak reference is a reference that is still owned by the garbage-collection and is released there when necessary.
- The self reference in MkObjectS::self only serves as a hint that there might be an object in the target language that resolves to
NULLif necessary.
- created with:
IntPtr GCHandle.Alloc(this,GCHandleType.Weak) - resolved with
object? GCHandle.FromIntPtr(self).Target(null return possible) - released with:
GCHandle.FromIntPtr(self).Free()
- The self reference in MkObjectS::self only serves as a hint that there might be an object in the target language that resolves to
- strong
- A strong reference is a reference that is owned by the Programming-Language-Micro-Kernel (PLMK) and is not released by the garbage-collection.
- The reference in MkObjectS::self is guaranteed to resolve to a valid object in the Target-Programming-Language (TPL).
- The object is passed into garbage-collection by releasing the strong reference.
- created with:
IntPtr GCHandle.Alloc(this) - resolved with
object GCHandle.FromIntPtr(self).Target(null return not possible) - released with:
GCHandle.FromIntPtr(self).Free()
- PLMK is able to deal with a weak or a strong reference.
- By default the weak reference is used, a strong reference is activated with the compile-switch
META_STRONG_REF
PERFORMANCE
I only compare the performance of C# with the performance of Java as a direct competitor.
call an empty callback from C
- CS code
- private void NTHT () {}
- CS result
call: 'NHI1_HOME/performance/performance.bash' 'pr' '--send-nothing' '--sec' '4' '_pipe' setup=perf-release > feature=cc_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ .../perf-release/inst/sbin/cc/x86_64-suse-linux-gnu-perfserver C> {x86_64-suse-linux-gnu-perfclient:pid(25468):tid(0x7f6c027a12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6c0278c440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25468):tid(0x7f6c027a12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6c0278c440):statistics }: --send-nothing : 535975.8 [ 2144169 / 4.000496 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25468):tid(0x7f6c027a12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6c0278c440):PerfClientExec }: end: ---------------------------------------- > feature=c_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfserver C> {x86_64-suse-linux-gnu-perfclient:pid(25484):tid(0x7f9b869c62c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f9b869b1440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25484):tid(0x7f9b869c62c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f9b869b1440):statistics }: --send-nothing : 543397.0 [ 2173706 / 4.000217 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25484):tid(0x7f9b869c62c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f9b869b1440):PerfClientExec }: end: ---------------------------------------- > feature=cs_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ /usr/bin/mono .../perf-release/inst/share/NHI1/perfserver.exe C> {x86_64-suse-linux-gnu-perfclient:pid(25501):tid(0x7fc1251932c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fc12517e440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25501):tid(0x7fc1251932c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fc12517e440):statistics }: --send-nothing : 429123.1 [ 1716758 / 4.000619 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25501):tid(0x7fc1251932c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fc12517e440):PerfClientExec }: end: ---------------------------------------- > feature=jv_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ /usr/bin/java -jar .../perf-release/inst/share/NHI1/perfserver.jar C> {x86_64-suse-linux-gnu-perfclient:pid(25520):tid(0x7f554c7812c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f554c76c440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25520):tid(0x7f554c7812c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f554c76c440):statistics }: --send-nothing : 472206.4 [ 1888940 / 4.000242 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25520):tid(0x7f554c7812c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f554c76c440):PerfClientExec }: end: ---------------------------------------- > feature=py_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/python3 .../perf-release/inst/sbin/py/x86_64-suse-linux-gnu-perfserver.py C> {x86_64-suse-linux-gnu-perfclient:pid(25556):tid(0x7f778c57b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f778c566440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25556):tid(0x7f778c57b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f778c566440):statistics }: --send-nothing : 499006.3 [ 1996153 / 4.000256 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25556):tid(0x7f778c57b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f778c566440):PerfClientExec }: end: ---------------------------------------- > feature=rb_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/ruby .../perf-release/inst/sbin/rb/x86_64-suse-linux-gnu-perfserver.rb C> {x86_64-suse-linux-gnu-perfclient:pid(25575):tid(0x7f6a98ed12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6a98ebc440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25575):tid(0x7f6a98ed12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6a98ebc440):statistics }: --send-nothing : 471033.9 [ 1884287 / 4.000321 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25575):tid(0x7f6a98ed12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f6a98ebc440):PerfClientExec }: end: ---------------------------------------- > feature=tcl_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --send-nothing --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/tclsh8.6 .../perf-release/inst/sbin/tcl/x86_64-suse-linux-gnu-perfserver.tcl C> {x86_64-suse-linux-gnu-perfclient:pid(25593):tid(0x7f516b7f02c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f516b7db440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25593):tid(0x7f516b7f02c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f516b7db440):statistics }: --send-nothing : 414436.1 [ 1657972 / 4.000549 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25593):tid(0x7f516b7f02c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f516b7db440):PerfClientExec }: end: ----------------------------------------
- analysis
- Java code is 5% faster
- C# and Java ar now slower than open-source C, C++ and Python
- C# is on the level as TCL
- The reason why Java is 5% faster than C# is the API connection. Java has a C connection and C# does not.
- The C# profiler check:
> Nhi1Exec -r=uds --extra-args='--profile=log:calls' perfserver.cs > Nhi1Exec -r=uds perfclient.c
- create the results:
... Method call summary Total(ms) Self(ms) Calls Method name ... Method call summary Total(ms) Self(ms) Calls Method name 1556 132 155628 (wrapper native-to-managed) csmqmsgque.MqContextC:ServiceCall (intptr,intptr,intptr) 1423 303 155628 csmqmsgque.MqContextC:ServiceCall (intptr,intptr,intptr) 626 95 155653 System.Runtime.InteropServices.GCHandle:FromIntPtr (intptr) 530 244 155653 System.Runtime.InteropServices.GCHandle:op_Explicit (intptr) 258 147 155637 System.Runtime.InteropServices.GCHandle:get_Target () 143 95 155653 System.Runtime.InteropServices.GCHandle:.ctor (intptr) 138 92 155628 csmkkernel.Mk:CsRuntimeGetErrorCode (intptr) 91 91 311306 intptr:op_Explicit (intptr) 90 90 311355 intptr:op_Equality (intptr,intptr) 63 63 155637 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:GetTarget (int) 53 53 155653 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:CheckCurrentDomain (int) 50 50 155626 example.perfserver:NTHT () 46 46 155637 System.Runtime.InteropServices.GCHandle:get_IsAllocated () 45 45 155638 intptr:ToPointer () ------------------------------------------------------- SUM 1546 >=155628 ... - Only the values with calls >= 155628 count for the performance comparison and there it can be seen that the three calls that represent the native implementation:
1556 132 155628 (wrapper native-to-managed) csmqmsgque.MqContextC:ServiceCall (intptr,intptr,intptr) 63 63 155637 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:GetTarget (int) 53 53 155653 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:CheckCurrentDomain (int) - are responsible for ~10% of the load and the actual "work":
50 50 155626 example.perfserver:NTHT ()
- only takes ~3% of the processing time.
- create the results:
call an MkBufferListC callback from C
- CS code
- private void BFLT () {var bfl = MkBufferListC.CreateTLS( "perfserver-BFLT" );while (ReadItemExists()) {bfl.AppendBUF(ReadBUF());}SendSTART();var size = bfl.Size();for (int i=0; i<size; i++) {SendBUF(bfl.IndexGet(i));};SendRETURN();}
- In the entire test cycle, no new objects are created. MkBufferListCreateTLS creates the MkBufferListC as thread-local and always returns the pale object with the identifier "perfserver-BFLT".
- This applies not only to the bfl but also to the MkBufferC contained therein, which are retrieved with MkBufferListIndexGet.
- CS result
call: 'NHI1_HOME/performance/performance.bash' 'pr' '--bfl' '--sec' '4' '_pipe' setup=perf-release > feature=cc_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ .../perf-release/inst/sbin/cc/x86_64-suse-linux-gnu-perfserver C> {x86_64-suse-linux-gnu-perfclient:pid(25656):tid(0x7f47b64432c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f47b642e440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25656):tid(0x7f47b64432c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f47b642e440):statistics }: --bfl : 73475.1 [ 293901 / 4.000010 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25656):tid(0x7f47b64432c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f47b642e440):PerfClientExec }: end: ---------------------------------------- > feature=c_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfserver C> {x86_64-suse-linux-gnu-perfclient:pid(25671):tid(0x7f88486352c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8848620440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25671):tid(0x7f88486352c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8848620440):statistics }: --bfl : 86672.4 [ 346690 / 4.000003 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25671):tid(0x7f88486352c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8848620440):PerfClientExec }: end: ---------------------------------------- > feature=cs_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ /usr/bin/mono .../perf-release/inst/share/NHI1/perfserver.exe C> {x86_64-suse-linux-gnu-perfclient:pid(25688):tid(0x7f515e38d2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f515e378440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25688):tid(0x7f515e38d2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f515e378440):statistics }: --bfl : 59200.3 [ 236802 / 4.000014 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25688):tid(0x7f515e38d2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f515e378440):PerfClientExec }: end: ---------------------------------------- > feature=jv_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ /usr/bin/java -jar .../perf-release/inst/share/NHI1/perfserver.jar C> {x86_64-suse-linux-gnu-perfclient:pid(25707):tid(0x7fac513c12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fac513ac440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25707):tid(0x7fac513c12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fac513ac440):statistics }: --bfl : 72401.4 [ 289606 / 4.000003 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25707):tid(0x7fac513c12c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fac513ac440):PerfClientExec }: end: ---------------------------------------- > feature=py_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/python3 .../perf-release/inst/sbin/py/x86_64-suse-linux-gnu-perfserver.py C> {x86_64-suse-linux-gnu-perfclient:pid(25742):tid(0x7f56411152c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f5641100440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25742):tid(0x7f56411152c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f5641100440):statistics }: --bfl : 64534.5 [ 258138 / 4.000003 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25742):tid(0x7f56411152c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f5641100440):PerfClientExec }: end: ---------------------------------------- > feature=rb_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/ruby .../perf-release/inst/sbin/rb/x86_64-suse-linux-gnu-perfserver.rb C> {x86_64-suse-linux-gnu-perfclient:pid(25759):tid(0x7f09c72062c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f09c71f1440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25759):tid(0x7f09c72062c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f09c71f1440):statistics }: --bfl : 66952.9 [ 267812 / 4.000003 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25759):tid(0x7f09c72062c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f09c71f1440):PerfClientExec }: end: ---------------------------------------- > feature=tcl_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --bfl --sec 4 @ /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/tclsh8.6 .../perf-release/inst/sbin/tcl/x86_64-suse-linux-gnu-perfserver.tcl C> {x86_64-suse-linux-gnu-perfclient:pid(25777):tid(0x7f2e9d33b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f2e9d326440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25777):tid(0x7f2e9d33b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f2e9d326440):statistics }: --bfl : 45807.1 [ 183229 / 4.000013 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25777):tid(0x7f2e9d33b2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f2e9d326440):PerfClientExec }: end: ----------------------------------------
- analysis
- C# is significantly slower than its competitors, special Java is ~20% faster than C#
- The C# profiler check:
> Nhi1Exec -r=uds --extra-args='--profile=log:calls' perfserver.cs > Nhi1Exec -r=uds perfclient.c
- create the results:
Method call summary Total(ms) Self(ms) Calls Method name 1830 7 7023 (wrapper native-to-managed) csmqmsgque.MqContextC:ServiceCall (intptr,intptr,intptr) 1823 15 7023 csmqmsgque.MqContextC:ServiceCall (intptr,intptr,intptr) 1755 69 7021 example.perfserver:BFLT () 704 63 70211 csmkkernel.MkBufferC:MkBufferC_ObjNew (intptr) 685 118 77240 csmkkernel.MkObjectC:atomObjNew (System.Type,System.Reflection.ConstructorInfo,intptr) 525 75 35106 csmqmsgque.MqContextC:ReadBUF () 523 75 35105 csmkkernel.MkBufferListC:IndexGet (int) 332 51 84273 System.Runtime.InteropServices.GCHandle:FromIntPtr (intptr) 285 189 323015 csmkkernel.MkObjectC:get_hdl () 280 125 84273 System.Runtime.InteropServices.GCHandle:op_Explicit (intptr) 204 73 35106 csmqmsgque.MqContextC:SendBUF (csmkkernel.MkBufferC) 166 166 561783 intptr:op_Equality (intptr,intptr) 143 53 35105 csmkkernel.MkBufferListC:AppendBUF (csmkkernel.MkBufferC) 135 77 84257 System.Runtime.InteropServices.GCHandle:get_Target () 114 68 77242 csmkkernel.Mk:CsGetSelfPtr (intptr) 100 11 7021 csmkkernel.MkBufferListC:CreateTLS (string,bool) 91 37 42126 csmqmsgque.MqContextC:ReadItemExists () 75 49 84273 System.Runtime.InteropServices.GCHandle:.ctor (intptr) 70 6 7021 csmkkernel.MkBufferListC:MkBufferListC_ObjNew (intptr) 62 41 70212 csmkkernel.MkObjectC:getOBJ (string,csmkkernel.MkObjectC) 52 52 168546 intptr:op_Explicit (intptr) 48 48 161530 csmkkernel.MkObjectC:get_mkrt () 46 46 154488 intptr:op_Inequality (intptr,intptr) 45 13 7022 csmqmsgque.MqContextC:SendRETURN () 44 35 119376 csmkkernel.MkErrorC:Check (intptr,csmkkernel.MkErrorE) 33 12 7022 csmqmsgque.MqContextC:SendSTART () 32 32 84257 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:GetTarget (int) 28 28 84273 (wrapper managed-to-native) System.Runtime.InteropServices.GCHandle:CheckCurrentDomain (int) 25 25 84264 intptr:ToPointer () 25 25 84257 System.Runtime.InteropServices.GCHandle:get_IsAllocated () 20 8 7021 csmkkernel.MkBufferListC:Size () 16 16 42126 (wrapper managed-to-native) csmqmsgque.Mq:MqReadItemExists (intptr) 16 16 35105 (wrapper managed-to-native) csmkkernel.Mk:MkBufferListAppendBUF (intptr,intptr,intptr) 15 15 7022 (wrapper managed-to-native) csmqmsgque.Mq:MqSendRETURN (intptr,intptr) 15 15 35106 (wrapper managed-to-native) csmqmsgque.Mq:MqSendBUF (intptr,intptr,intptr) 15 15 35106 (wrapper managed-to-native) csmqmsgque.Mq:MqReadBUF (intptr,intptr,intptr&) 14 14 35105 (wrapper managed-to-native) csmkkernel.Mk:MkBufferListIndexGet (intptr,intptr,int,intptr&) 10 4 7053 System.Runtime.InteropServices.Marshal:StringToHGlobalAnsi (string) - no significant reason for the C# overhead
- create the results:
compare CS versa JV in pipe mode
- results
call: 'NHI1_HOME/performance/performance.bash' 'pr' '--all-performance' '--sec' '4' '^(cs|jv)_pipe' setup=perf-release > feature=cs_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --all-performance --sec 4 @ /usr/bin/mono .../perf-release/inst/share/NHI1/perfserver.exe C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --send-nothing : 443287.0 [ 1773225 / 4.000174 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --send : 271675.3 [ 1086758 / 4.000210 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --send-and-callback : 154799.1 [ 619379 / 4.001178 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --send-and-wait : 74855.9 [ 299424 / 4.000005 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --parent : 21.7 [ 87 / 4.009635 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --child : 16513.6 [ 66055 / 4.000048 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --bus : 57408.0 [ 229633 / 4.000018 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):statistics }: --bfl : 60758.7 [ 243035 / 4.000000 ] C> {x86_64-suse-linux-gnu-perfclient:pid(25860):tid(0x7f83524872c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f8352472440):PerfClientExec }: end: ---------------------------------------- > mv 'NHI1_BUILD/performance/temp.perf.perf-release.cs_pipe' 'NHI1_HOME/performance/gen/x86_64-suse-linux-gnu/perf-release/cs_pipe.perf' > feature=jv_pipe > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --all-performance --sec 4 @ /usr/bin/java -jar .../perf-release/inst/share/NHI1/perfserver.jar C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --send-nothing : 476037.0 [ 1904387 / 4.000502 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --send : 298682.3 [ 1194774 / 4.000150 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --send-and-callback : 164119.9 [ 656634 / 4.000942 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --send-and-wait : 78644.3 [ 314578 / 4.000011 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --parent : 21.2 [ 85 / 4.000181 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --child : 19651.8 [ 78608 / 4.000040 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --bus : 70445.0 [ 281780 / 4.000002 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):statistics }: --bfl : 71173.7 [ 284695 / 4.000002 ] C> {x86_64-suse-linux-gnu-perfclient:pid(26147):tid(0x7fa5393a52c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fa539390440):PerfClientExec }: end: ---------------------------------------- > mv 'NHI1_BUILD/performance/temp.perf.perf-release.jv_pipe' 'NHI1_HOME/performance/gen/x86_64-suse-linux-gnu/perf-release/jv_pipe.perf'
- analysis
compare parent versa child startup in thread mode
- results
call: 'NHI1_HOME/performance/performance.bash' 'pr' '--parent' '--child' '--sec' '4' '_uds_thread' setup=perf-release > feature=cc_uds_thread > .../perf-release/inst/sbin/cc/x86_64-suse-linux-gnu-perfserver --uds --file ./socket.uds.0 --thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --parent --child --sec 4 --uds --file ./socket.uds.0 C> {x86_64-suse-linux-gnu-perfclient:pid(28188):tid(0x7f4adc5902c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4adc57b440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(28188):tid(0x7f4adc5902c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4adc57b440):statistics }: --parent : 10330.5 [ 41324 / 4.000177 ] C> {x86_64-suse-linux-gnu-perfclient:pid(28188):tid(0x7f4adc5902c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4adc57b440):statistics }: --child : 36505.2 [ 146021 / 4.000001 ] C> {x86_64-suse-linux-gnu-perfclient:pid(28188):tid(0x7f4adc5902c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4adc57b440):PerfClientExec }: end: ---------------------------------------- > feature=cs_uds_thread > /usr/bin/mono .../perf-release/inst/share/NHI1/perfserver.exe --uds --file ./socket.uds.1 --thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --parent --child --sec 4 --uds --file ./socket.uds.1 C> {x86_64-suse-linux-gnu-perfclient:pid(69575):tid(0x7f4ef20762c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4ef2061440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(69575):tid(0x7f4ef20762c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4ef2061440):statistics }: --parent : 11054.7 [ 44220 / 4.000118 ] C> {x86_64-suse-linux-gnu-perfclient:pid(69575):tid(0x7f4ef20762c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4ef2061440):statistics }: --child : 21754.1 [ 87017 / 4.000028 ] C> {x86_64-suse-linux-gnu-perfclient:pid(69575):tid(0x7f4ef20762c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f4ef2061440):PerfClientExec }: end: ---------------------------------------- > feature=c_uds_thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfserver --uds --file ./socket.uds.2 --thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --parent --child --sec 4 --uds --file ./socket.uds.2 C> {x86_64-suse-linux-gnu-perfclient:pid(69613):tid(0x7fb03340f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fb0333fa440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(69613):tid(0x7fb03340f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fb0333fa440):statistics }: --parent : 12347.8 [ 49393 / 4.000144 ] C> {x86_64-suse-linux-gnu-perfclient:pid(69613):tid(0x7fb03340f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fb0333fa440):statistics }: --child : 37620.2 [ 150481 / 4.000005 ] C> {x86_64-suse-linux-gnu-perfclient:pid(69613):tid(0x7fb03340f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7fb0333fa440):PerfClientExec }: end: ---------------------------------------- > feature=jv_uds_thread > /usr/bin/java -jar .../perf-release/inst/share/NHI1/perfserver.jar --uds --file ./socket.uds.3 --thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --parent --child --sec 4 --uds --file ./socket.uds.3 C> {x86_64-suse-linux-gnu-perfclient:pid(46228):tid(0x7f78085882c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f7808573440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(46228):tid(0x7f78085882c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f7808573440):statistics }: --parent : 13961.1 [ 55846 / 4.000111 ] C> {x86_64-suse-linux-gnu-perfclient:pid(46228):tid(0x7f78085882c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f7808573440):statistics }: --child : 19807.1 [ 79229 / 4.000040 ] C> {x86_64-suse-linux-gnu-perfclient:pid(46228):tid(0x7f78085882c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f7808573440):PerfClientExec }: end: ---------------------------------------- > feature=tcl_uds_thread > /home/dev1usr/ext/x86_64-suse-linux-gnu/release/bin/tclsh8.6 .../perf-release/inst/sbin/tcl/x86_64-suse-linux-gnu-perfserver.tcl --uds --file ./socket.uds.4 --thread > .../perf-release/inst/sbin/c/x86_64-suse-linux-gnu-perfclient --timeout 2 --parent --child --sec 4 --uds --file ./socket.uds.4 C> {x86_64-suse-linux-gnu-perfclient:pid(46269):tid(0x7f726410f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f72640fa440):PerfClientExec }: start ------------------------ : result [ count / sec ] C> {x86_64-suse-linux-gnu-perfclient:pid(46269):tid(0x7f726410f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f72640fa440):statistics }: --parent : 38.5 [ 154 / 4.000189 ] C> {x86_64-suse-linux-gnu-perfclient:pid(46269):tid(0x7f726410f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f72640fa440):statistics }: --child : 23844.6 [ 95379 / 4.000020 ] C> {x86_64-suse-linux-gnu-perfclient:pid(46269):tid(0x7f726410f2c0):L:dlv(0):ctxId( 0):rc(0):ctx(0x7f72640fa440):PerfClientExec }: end: ----------------------------------------
- analysis
- The values were determined for a single processor (–wrk 1) environment.
- The real performance can be multiplied by the number of physical processors…
- The maximum achievable number of new sessions (
--parent) with thread startup and the new thread-pool functionality is limited by the time for creating a new context (--child). - A Xeon E3-1275 V2 with 4 physical processors can therefore process max 150,000 sessions per second.
- calculation for a C application server:
4 processors x 37,000 ctx/sec = ~150,000 ctx/sec.
- calculation for a C application server:
- The maximum achievable number of new sessions (
- C# is significant slower the Java in thread startup even if a thread-pool is used.
Generated on Tue Dec 31 2024 00:11:39 for NHI1 by
1.12.0