While julia is opening, it's pegged at around 99% CPU usage. It drops to 
something very low after startup, though.

On Wednesday, August 26, 2015 at 4:02:35 PM UTC-4, Elliot Saba wrote:
>
> Does `top` show that Julia is taking up a huge amount of CPU or memory?
> -E
>
> On Wed, Aug 26, 2015 at 11:49 AM, Chris <[email protected] <javascript:>
> > wrote:
>
>> That gives "/home/cbinz/julia/usr/bin/../lib/julia/sys.so"
>>
>> On Wednesday, August 26, 2015 at 2:47:06 PM UTC-4, Elliot Saba wrote:
>>>
>>> After Julia is loaded, run the following command and tell us what it 
>>> prints out; it will give us the same information as what Yichao is talking 
>>> about.
>>>
>>> filter( x -> contains(x, "sys.$(Sys.dlext)"), Sys.dllist())
>>> -E
>>>
>>> On Wed, Aug 26, 2015 at 11:43 AM, Chris <[email protected]> wrote:
>>>
>>>> I have a sys.so in ~/julia/usr/lib/julia/, but I'm not sure how to run 
>>>> fuser or lsof to do what you're asking.
>>>>
>>>> On Wednesday, August 26, 2015 at 2:32:32 PM UTC-4, Yichao Yu wrote:
>>>>
>>>>> On Wed, Aug 26, 2015 at 2:20 PM, Chris <[email protected]> wrote: 
>>>>> > ipython is not installed, but `python` seems to run just fine. 
>>>>> > 
>>>>> > Nothing jumps out at me as taking very long with `strace julia`. 
>>>>>
>>>>> Maybe do a lsof/fuser on the sys.so (should be in /usr/lib/julia/ or 
>>>>> similar directory) to see if it is actually loaded by julia? (Or if it 
>>>>> exists at all) 
>>>>>
>>>>> > 
>>>>> > On Wednesday, August 26, 2015 at 2:14:48 PM UTC-4, Elliot Saba 
>>>>> wrote: 
>>>>> >> 
>>>>> >> Do other interpreters such as `ipython` run slowly? 
>>>>> >> 
>>>>> >> If you run `strace julia`, it will print out system calls as they 
>>>>> execute. 
>>>>> >> Are there certain syscalls that are taking a long amount of time?  
>>>>> E.g. does 
>>>>> >> it freeze for a long time on an lstat(), or a read(), or something? 
>>>>> >> -E 
>>>>> >> 
>>>>> >> On Wed, Aug 26, 2015 at 11:09 AM, Chris <[email protected]> 
>>>>> wrote: 
>>>>> >>> 
>>>>> >>> Nope, not NFS-mounted. 
>>>>> >>> 
>>>>> >>> 
>>>>> >>> On Wednesday, August 26, 2015 at 2:07:14 PM UTC-4, John Gibson 
>>>>> wrote: 
>>>>> >>>> 
>>>>> >>>> Run "df ~". That'll tell you where the file system containing 
>>>>> your home 
>>>>> >>>> directory is mounted. If it says "nfs:/...", it's NFS-mounted. 
>>>>> >>>> 
>>>>> >>>> John 
>>>>> >>>> 
>>>>> >>>> On Wednesday, August 26, 2015 at 11:12:48 AM UTC-4, Chris wrote: 
>>>>> >>>>> 
>>>>> >>>>> Hello, 
>>>>> >>>>> 
>>>>> >>>>> I recently got access to a new Linux machine, and I've been 
>>>>> trying to 
>>>>> >>>>> run some of my code there. I tried downloading a binary and 
>>>>> using that, but 
>>>>> >>>>> after I saw the performance issues, I built Julia from source, 
>>>>> and the 
>>>>> >>>>> issues persist. First: 
>>>>> >>>>> 
>>>>> >>>>>  julia> @time versioninfo() 
>>>>> >>>>> Julia Version 0.3.12-pre+5 
>>>>> >>>>> Commit 24138e7 (2015-08-20 17:19 UTC) 
>>>>> >>>>> Platform Info: 
>>>>> >>>>>   System: Linux (x86_64-linux-gnu) 
>>>>> >>>>>   CPU: Intel(R) Xeon(R) CPU E5-4650 0 @ 2.70GHz 
>>>>> >>>>>   WORD_SIZE: 64 
>>>>> >>>>>   BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY 
>>>>> Sandybridge) 
>>>>> >>>>>   LAPACK: libopenblas 
>>>>> >>>>>   LIBM: libopenlibm 
>>>>> >>>>>   LLVM: libLLVM-3.3 
>>>>> >>>>> elapsed time: 40.533040322 seconds (142031648 bytes allocated, 
>>>>> 2.89% gc 
>>>>> >>>>> time) 
>>>>> >>>>> 
>>>>> >>>>> You can see that is quite a long time for the command to run (on 
>>>>> my 
>>>>> >>>>> Windows machine, it takes about 1.3 seconds). Startup itself is 
>>>>> quite slow, 
>>>>> >>>>> and even typing input immediately after startup is slow (the 
>>>>> characters take 
>>>>> >>>>> a few seconds just to show up). Loading a module that I use a 
>>>>> lot on Windows 
>>>>> >>>>> takes about 20 seconds, but loading that same module on the 
>>>>> Linux machine 
>>>>> >>>>> takes almost 9 minutes! 
>>>>> >>>>> 
>>>>> >>>>> Obviously something is wrong here, and I'm stumped. Any ideas? 
>>>>> >>>>> 
>>>>> >>>>> Thanks, 
>>>>> >>>>> Chris 
>>>>> >> 
>>>>> >> 
>>>>> > 
>>>>>
>>>>
>>>
>

Reply via email to