(Uncompleted document in some ones, so please wait a minute)
Introduction to how to use AWS Ec2 instance through JAIST CLOUD
ISC provides kindly these with purpose to acceralate our research and to have us acquire sophisticated information processing skill on educational scenario. (We can use freely but) Note that we should notice we have a duty to avoid "over-use" when we use it any time because resource cost... (when you take a rest (e.g go to bed, have a food and so on...), computer also hope to do that 😴)
We suggest definitive standard time for switch-off as 1 day, it means we switch off if we do not use over 1 day a last use (e.g. attend to conference, take a trip etc...)
-
Intro
-
Env
-
Setting basic environment
sudo yum update- CUDA sample test, comple 1
- In a case you download cudnn in Windows, you need to change its identifier into arbitary one. 2
- Amazon Linux AMI like Redhat distribution, p2 EC2
- CPU Xeon E5-2686v4 (Broadwell) *4
- GPU NVIDIA K80 *1
- mem 61?(GiB) (CPU), 12(GiB) (GPU)
- See more jp site here https://aws.amazon.com/jp/ec2/instance-types/ ,
-
For general document of how to use on JCS, we can view here(en), here(jp)
-
import .pem file (certificate)
-
login
-
mount in my dir
-
chmod filesystem(mounted directory), chown? (which proper to permission)
- /etc/??? (vi edit ___ ) to fix mnt after switching-off
-
cuda /.bash_proofile ==> note that data position,,,, after that, relogin to active roader it
-
()cuDNN (note that data position,,,, scp)
-
pyenv in data ~ miniconda ~
-
conda config to data http://nekoyukimmm.hatenablog.com/entry/2016/03/27/173030
-
tf (sample code to check work on gpu)
-
()jy port
- CUDA (already installed 7.5)
- CuDNN(need to download in local and then pass through it to remote with matched version for 7.5) 6.0
sudo su -- We need to consider to install and expand cuDNN tgz in enough inode (valume) file.
- Conda
- Deep learning tools
- Tensorflow, Chainer, Keras
- Mxnet
- Spyder (remote)
- (Atom, hidrogen? conda, jupyter, IDE)
- Jupyter remote, lab, theme - jupyter lab, and color (font "Recty Dimished"). (extension is under the development) (Nodejs kernel still not)
- We need to check about specific version in GPU, to use it.
mount extended HDD- inode issue (how can we resolve it? uses xft? ext4(how can check), root?user? => can file mnt folder or expand inodes?)
static IP(fixed initially by defalt in Jaist setting)- In RHEL(or CentOS)-like AWS Linux, any equivalent for libcupti-dev library: https://stackoverflow.com/questions/45348033/tensorflow-installation-on-centos7-libcupti-dev-equivalent , rpm ===> Hint: libcupti /usr/local/cuda-8.0/extras/CUPTI/lib64
- Making sample code, data analysis, deep learning (but mainly around my theme... see more memo)
- Docker
- qiita
- tf p2 (CUDA, CuDNN-From-scp)(miniconda, jupyter)(security group for jupyter needed? env,path?) https://qiita.com/hamaguchitm/items/269df3662a09f46277c7
- (this aws linux ami does not work? DON'T USE anaCONDA) https://qiita.com/halhorn/items/361008b19b4fcfd618d6
- tf
- (CuDNN acceralate calcu, DETAIL FILE MNG) https://qiita.com/h860a/items/294262d98e1223008252
- anaconda+jupyter(don't use sec gr?) simple https://qiita.com/Salinger/items/c7b87d7000e48be6ebe2
- mxnet
- web
- tf p2 (CUDA, CuDNN-From-scp)(miniconda, jupyter)(security group for jupyter needed? env,path?) https://qiita.com/hamaguchitm/items/269df3662a09f46277c7