shafiee01

4 ms·

1.
by shafiee01 11y ago
This is really useful; I have been recently working on a huge c++ project and I had the exact same problem you mentioned with large vectors. The way I handled this problem was to wrap vector in a class which provides locking mechanism for e
2.
by shafiee01 12y ago
Yes, I will talk about that with my supervisor.
3.
by shafiee01 12y ago
If you flush a file it will be in the backend; Therefore, even if a node goes down, another node will be responsible for that. There will be obviously a delay for the recovery. In POSIX if your buffer is not flushed there is no guarantee th
4.
by shafiee01 12y ago
I wish if I could :) but right now in our lab I don't have access to such equipments. Honestly, that was the plan because using 10/40G networks will significantly reduce the delay in remote operations as well as increasing the thr
5.
by shafiee01 12y ago
Yes, I plan to run experiments with different numbers of threads/files and many other factors. Right know I only have access to a cluster of 7 old machines. That's the best I can do and for the hadoop case I used all of the machi
6.
by shafiee01 12y ago
1- This is just a preliminary evaluation. I just made the repository public 3 hours ago. I am planning so many experiments including the ones you mentioned. 2- As I said more benchmark including multiple files with iozone, postmark are comi
7.
by shafiee01 12y ago
1-Well if you don't want to use a backend storage what else do you expect? 2-It will flush as soon as you call flush or close your file. It's not necessarily disk and can be any configured backend (right now only swift is supporte
8.
An in-memory distributed Posix-compatible file system(bshafiee.github.io)
31 points by shafiee01 12y ago|13 comments