- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
Recently, the compilation time of our programs has slow down dramatically. For a very simple source :
@drhpcmss % time ifort -c -fpp hsbib/normR.f90
real 0m12.450s
user 0m0.048s
sys 0m0.084s
We believe that this comes from the license check, but running the compilation with $INTEL_LICENSE_FILE pointing explicitely to a folder containing a license file (which itself point to a server) does not help. Below is the results with $INTEL_LMD_DEBUG=1 (most of the time is between "INTEL_LMD: checkoutfilter: returns ACCEPT" and "INTEL_LMD: checkout: checkout call returns rc 0") .
Any suggestion that could solve the issue ?
Regards,
Guillaume
drhpcmss % time env INTEL_LMD_DEBUG=1 ifort -c -fpp hsbib/normR.f90
INTEL_LMD: flex_set_license_path: called, path=</opt/intel/composer_xe_2013_sp1.4.211/bin/intel64>
INTEL_LMD: flex_set_platform: called, plat_name=<i86_r>
INTEL_LMD: checkout: entered at start_time: 1412689314 (UTC) Tue Oct 7 13:41:54 2014
INTEL_LMD: checkout: build options:
INTEL_LMD: checkout: build date: 2014.0806
INTEL_LMD: checkout: build platform: Intel(R) 64 architecture; Linux*
INTEL_LMD: checkout: checkout platform: i86_r (IA-32 architecture; Linux*)
INTEL_LMD: checkout: *Other brands and names are the property of their respective owners.
INTEL_LMD: checkout: FLEXlm version: 11.12
INTEL_LMD: checkout: source version: 1.132 $ Intel(R) Software Licensing release 2.3 | FlexNet Publisher* version 11.12.0 $
INTEL_LMD: checkout: FLEX_MAX_LICENSE_TRIES: 0
INTEL_LMD: checkout: FLEX_USE_NOWAIT_AND_SPIN: enabled
INTEL_LMD: checkout: FLEX_SPIN_SLEEP_SECONDS: 30
INTEL_LMD: checkout: heartbeats: 60 seconds (automatic)
INTEL_LMD: checkout: server retry: 60 seconds
INTEL_LMD: checkout: FLEX_CHECKOUT_FILTER: enabled
INTEL_LMD: checkout: FLEX_CLEAN_PATHS: disabled
INTEL_LMD: checkout: FLEX_CLOCK_ROLLBACK_CHECK: disabled
INTEL_LMD: checkout: FLEX_APP_SPECIFIC: disabled
INTEL_LMD: checkout: user options:
INTEL_LMD: checkout: feature name: FCompL
INTEL_LMD: checkout: opts: { FLEX_OPT_QUIET FLEX_OPT_QUIET_EXPIRE_EVAL }
INTEL_LMD: checkout: try_count: 0
INTEL_LMD: checkout: retry_interval: 30
INTEL_LMD: checkout: linger_time: 0
INTEL_LMD: checkout: environment:
INTEL_LMD: checkout: INTEL_LICENSE_FILE </opt/intel/composer_xe_2013_sp1.4.211/licenses:/opt/intel/licenses:/homes/gde-hauteclocque/intel/licenses:/opt/intel/composer_xe_2013_sp1.4.211/Licenses:/homes/gde-hauteclocque/intel/licenses:/opt/intel/licenses:/Users/Shared/Library/Application Support/Intel/Licenses:/opt/intel/composer_xe_2013_sp1.4.211/bin/intel64>
INTEL_LMD: checkout: LM_LICENSE_FILE <NULL>
INTEL_LMD: checkout: INTEL_LMD_DEBUG <1>
INTEL_LMD: checkout: license file(s) used were (in this order):
INTEL_LMD: ** 1. /opt/intel/composer_xe_2013_sp1.4.211/licenses
INTEL_LMD: ** 2. /opt/intel/licenses/USE_SERVER.lic
INTEL_LMD: ** 3. /homes/gde-hauteclocque/intel/licenses
INTEL_LMD: ** 4. /Users/Shared/Library/Application Support/Intel/Licenses
INTEL_LMD: ** 5. /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/*.lic
INTEL_LMD: checkout: linger_time set to 0 by default
INTEL_LMD: checkout: begin attempt(s) to checkout
INTEL_LMD: checkout: about to do checkout call
INTEL_LMD: checkoutfilter: called for feature: FCompL
INTEL_LMD: checkoutfilter: VENDOR_STRING: <SUPPORT=COM https://registrationcenter.intel.com>;
INTEL_LMD: checkoutfilter: next VENDOR_STRING keyword: SUPPORT=<COM>
INTEL_LMD: checkoutfilter: ignore: keyword SUPPORT is disabled for filtering
INTEL_LMD: checkoutfilter: next VENDOR_STRING keyword: https://registrationcenter.intel.com
INTEL_LMD: checkoutfilter: ignore: keyword https://registrationcenter.intel.com is disabled for filtering
INTEL_LMD: checkoutfilter: returns ACCEPT
INTEL_LMD: checkout: checkout call returns rc 0
INTEL_LMD: checkout: license now granted
INTEL_LMD: flex_config: called
INTEL_LMD: flex_config: at least one license has been granted
INTEL_LMD: flex_config: the most recently checked out feature was checked out successfully
INTEL_LMD: checkout: contents of particular license actually checked out:
INTEL_LMD: checkout: feature name: FCompL (INCREMENT line)
INTEL_LMD: checkout: serial number: SMSADNV27DB5
INTEL_LMD: checkout: license expires: perpetual license (does not expire)
INTEL_LMD: checkout: license maintenance expires: 2015.0926
INTEL_LMD: checkout: type of license: counted, 2 users maximum
INTEL_LMD: checkout: allowed platforms:
INTEL_LMD: checkout: i86_r (IA-32 architecture; Linux*)
INTEL_LMD: checkout: i86_re (IA-32 architecture; Linux*)
INTEL_LMD: checkout: it64_lr (IA-64 architecture; Linux*)
INTEL_LMD: checkout: it64_re (IA-64 architecture; Linux*)
INTEL_LMD: checkout: amd64_re (Intel(R) 64 architecture; Linux*)
INTEL_LMD: checkout: *Other brands and names are the property of their respective owners.
INTEL_LMD: checkout: VENDOR_STRING= string contents:
INTEL_LMD: checkout: SUPPORT=<COM>
INTEL_LMD: checkout: https://registrationcenter.intel.com
INTEL_LMD: flex_config: called
INTEL_LMD: flex_config: at least one license has been granted
INTEL_LMD: flex_config: the most recently checked out feature was checked out successfully
INTEL_LMD: flex_expire_days: called
INTEL_LMD: flex_config: called
INTEL_LMD: flex_config: at least one license has been granted
INTEL_LMD: flex_config: the most recently checked out feature was checked out successfully
INTEL_LMD: flex_expire_days: returns 3650000
INTEL_LMD: checkout: returns GRANTED
INTEL_LMD: flex_config: called
INTEL_LMD: flex_config: at least one license has been granted
INTEL_LMD: flex_config: the most recently checked out feature was checked out successfully
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_pre_fork: called
INTEL_LMD: flex_post_fork: called
INTEL_LMD: flex_return_license: called
INTEL_LMD: flex_return_license: considering feature name: FCompL
INTEL_LMD: flex_return_license: feature name needs checkin: FCompL
INTEL_LMD: flex_return_license: checkin succeeded for feature: FCompL
INTEL_LMD: flex_free_job: called
INTEL_LMD: flex_free_job: lc_new_job was called
INTEL_LMD: flex_return_license: done
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
The problem persists and make the compilation so slow (more than 4hours for the whole program) that the compiler is barely usable. A few hint that could help in helping us :
1- We do not observe the problem on windows (the license server for Linux and Windows is the same, and linux based)
2- On our linux system, with a few exception, the computers run with AMD processors. The few computer with intel processor does not show similar issue. Might that be relevant ?
Regards,
Guillaume
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I am not aware that the CPU brand has any significance here. We do have some similar complaints about license checkout times on Linux. I'll let the people investigating this know of the problem you are seeing.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The developers have asked that you try an experiment and replace the host name in the local license file (the one with USE SERVER) with the IP address of the license server. See if that makes a difference.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
NO, I don't think it is a licensing configuration problem.
The intel ifort compiler is installed on a DL365G7 HP Proliant server (hostname drhpcmpp ) :
fp2x@drhpcmpp:/tmp> df -kT /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
Sys. de fichiers Type 1K-blocs Util. Dispo. Uti% Monté sur
/dev/mapper/opt_vg-opt_lv ext3 61927420 58473500 2825648 96% /opt
fp2x@drhpcmpp:/tmp> time /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m12.315s
user 0m0.004s
sys 0m0.012s
Please note that there is no licensing error message.
fp2x@drhpcmpp:/tmp> cat /opt/intel/licenses/USE_SERVER.lic
SERVER drsvn.ho.bureauveritas.com ANY 27000
USE_SERVER
Now I try the same ifort invocation from another linux 64 system with /opt/intel directory automounted :
fp2x@drbuildbot:~$ time /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m0.550s
user 0m0.000s
sys 0m0.012s
fp2x@drbuildbot:~$ df -kT /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
Sys. fich. Type 1K-blocks Util. Disponible Uti% Monté sur
drhpcmpp:/opt/intel nfs4 61927424 58473472 2826240 96% /opt/intel
Depending on the time to mount the /opt/intel nfs4 file system the time of invocation varies, it can be as low as 0.125 second
fp2x@drbuildbot:~$ time /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m0.125s
user 0m0.004s
sys 0m0.008s
But on the very machine which holds the file system the delay is consistently always above 12.125 seconds
My explaination :
On the machine with a > 12 sec timeout delay
fp2x@drhpcmpp:/tmp> fgrep vendor_id /proc/cpuinfo
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
vendor_id : AuthenticAMD
On the no delay machine
fp2x@drbuildbot:~$ fgrep vendor_id /proc/cpuinfo
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
Now, let's try with an openSUSE 11.4 (x86_64) system (named linux02 ) :
fp2x@linux02:~> time /optl/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m0.102s
user 0m0.001s
sys 0m0.007s
fp2x@linux02:~> df -kT /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
Sys. fich. Type 1K-blocs Utilisé Dispo. Uti% Monté sur
drhpcmpp:/opt/intel/
nfs4 61927424 58473472 2826240 96% /optl/intel
fp2x@linux02:~> fgrep vendor_id /proc/cpuinfo
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
vendor_id : GenuineIntel
The automount is made on /optl as /opt is used by the system
I remove the sybolic link /opt/intel -> /optl/intel :
fp2x@linux02:~> sudo rm /opt/intel
fp2x@linux02:~> time /optl/intel/composer_xe_2013_sp1.4.211/bin/intel64/ifort
Error: A license for FCompL is not available (-76,61026,2).
License file(s) used were (in this order):
1. Trusted Storage
** 2. /optl/intel/composer_xe_2013_sp1.4.211/Licenses/USE_SERVER.lic
** 3. /home/fp2x/intel/licenses
** 4. /opt/intel/licenses
** 5. /Users/Shared/Library/Application Support/Intel/Licenses
** 6. /optl/intel/composer_xe_2013_sp1.4.211/bin/intel64/*.lic
Please visit http://software.intel.com/sites/support/ if you require technical assistance.
ifort: error #10052: could not checkout FLEXlm license
real 0m1.051s
user 0m0.010s
sys 0m0.020s
The 1 second duration to check the paths of licences seems reasonable
The systematic 12 seconds delay when the compiling system has AMD processors is unacceptable.
As the compilation of most source files is very quick, there is a two orders of magnitude performance lost in case of AMD64 CPUs.
The prejudice for our Research Department is considerable as we bought 6 DL165G7 ( 2x12 cores ), one DL365G7 ( 2x8 cores ) and a 1x8 cores DL365G7 (drhpcmpp) machines, 4 years ago.
It is the same Debian binaries on the DL165G7 machines and the drbuildbot system ( GenuineIntel CPU ). No compatibility problem of any sort, except with Intel Software products
Cordialement,
Regards,
Mit freundlichen Grüssen,
ﻢﻋ ﺖﺤﻳﺎﺘﻳ ﺎﻠﺧﺎﻠﺻﺓ
---
F. Petitjean
01 55 24 75 05
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have to think that there is something else at work other than the processor brand. We do not even look at this in our code and we use a third-party licensing library for the actual license checks. Thanks for the detailed information and I will forward that to the developers.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Trying with the last version
fp2x@drbuildbot:~$ time /opt/intel/composer_xe_2015.1.133/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m0.120s
user 0m0.000s
sys 0m0.008s
fp2x@drbuildbot:~$ time /opt/intel/composer_xe_2015.1.133/bin/intel64/icc
icc: command line error: no files specified; for help type "icc -help"
real 0m0.120s
user 0m0.004s
sys 0m0.004s
very easy to reproduce
fp2x@drbuildbot:~$ fgrep vendor_id /proc/cpuinfo | uniq
vendor_id : GenuineIntel
___________
fp2x@drhpcmpp:~> time /opt/intel/composer_xe_2015.1.133/bin/intel64/ifort
ifort: command line error: no files specified; for help type "ifort -help"
real 0m12.133s
user 0m0.012s
sys 0m0.008s
fp2x@drhpcmpp:~> time /opt/intel/composer_xe_2015.1.133/bin/intel64/icc
icc: command line error: no files specified; for help type "icc -help"
real 0m12.133s
user 0m0.000s
sys 0m0.020s
fp2x@drhpcmpp:~> fgrep vendor_id /proc/cpuinfo | uniq
vendor_id : AuthenticAMD
fp2x@drhpcmpp:~>
Please, why a penalty of twelve seconds ?
The AuthenticAMD AMD64 eight cores CPU has the Magny-Cours architecture
---
Cordialement,
Regards,
Mit freundlichen Grüssen,
ﻢﻋ ﺖﺤﻳﺎﺘﻳ ﺎﻠﺧﺎﻠﺻﺓ
---
F. Petitjean
01 55 24 75 05
Intel société américaine qui à la suite de pratiques anticoncurrentielles systématiques s'est vue infliger une amende record de 1060 millions € par la Commission Européenne.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
No answers yet - sorry.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I know it's been a long time, but I finally got an answer. The developers are reasonably certain that the issue is slow response from the DNS server used by the system. Even putting an IP address in the license file for the server location doesn't skip DNS lookups. As I indicated before, the CPU type is not relevant.
The recommendation I have received is to check your "nsswitch.conf" file and make sure it checks /etc/hosts first. Another thing to look at is resolv.conf and perhaps change the order of DNS servers to have a faster one come first.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
We were suffering the same problem. I used the same machine for both license server and client. The DNS setup is absolutely correct. But the checkout is very slow. The moment we disabled the DNS in /etc/resolve.cnf the checkout became fast.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
could it be possible that these two topics are related to each other?
'ifort.exe is slow on windows': https://software.intel.com/en-us/forums/intel-visual-fortran-compiler-for-windows/topic/598158
It's Windows and not Linux but it's the same license management bib (flexlm) used by Intel compilers and the description sounds similar.
Regards, Johannes
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I think separate issues are getting confused.
@Yi C - The OP was about slowness between two systems with different CPUs. It sounds like your issue relates to having DNS enabled. What happens if you reordered nssswitch.conf as Steve described? What release of the Parallel Studio XE are you using where you see the slow checkout, 2015 or 2016?
@Johannes - The slowness reported on Windows in the thread you cited is reportedly only occurring with the latest PSXE 2016 (16.0 compiler) and that it does not occur with the previous PSXE 2015 (15.0) compiler so that makes me believe these issues (the non-Intel CPU and DNS configuration) are not connected; however, we won't dismiss the possibility and will share Yi C's observation with our Developers.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I had a similar issue, I realized that the DNS query that was sometimes slow was an IPV6 query (AAAA) on the short hostname of the building host. I resolved it by putting a fake IPV6 entry in /etc/hosts, is there a way to avoid intel compiler issuing this name query ? And why is it issued in the first place, I don't get its purpose ?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
It's the licensing check that does the DNS lookup - I'm not sure exactly what it's doing here as it's part of the licensing solution we use. I have heard of other issues relating to DNS lookups slowing the compiler from starting.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I install CentOS 7.2 and had the same problem with the Intel compiler license delay. CentOS 7.2 enables ipv6 by default.
login1.hikari(2)# grub2-mkconfig -o /boot/efi/EFI/centos/grub.cfg
login1.hikari(3)# reboot

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page