| 1 |
Hi all, |
| 2 |
|
| 3 |
On Thu, 25 Feb 2016 22:03:59 +0000 (UTC) James wrote: |
| 4 |
> This smoking hot (many HPC scientist agree) distributed file |
| 5 |
> system will surely rock the cluster, container and Hi Performance |
| 6 |
> Computing worlds. [1] Now if I were only smart enough to get this |
| 7 |
> puppy into portage....... |
| 8 |
|
| 9 |
By the way, does anyone have any real performance comparison with |
| 10 |
Lustre? |
| 11 |
|
| 12 |
While it is good to have another solution available, I don't see |
| 13 |
any real benefits of FhgFS/BeeGFS compared to Lustre these days. |
| 14 |
At the time where FhgFS was created, Lustre indeed was unable to |
| 15 |
use multiple metadata servers, so this was a bottleneck. But now |
| 16 |
Lustre also supports distributed metadata, so they should on par in |
| 17 |
this matter. |
| 18 |
|
| 19 |
On the other hand, Lustre has much larger community (e.g. see |
| 20 |
TOP-500 list) and is much better tested (and even under such |
| 21 |
conditions it has problems in some corner cases). Thus I see no |
| 22 |
advantage in FhgFS for HPC setups. |
| 23 |
|
| 24 |
Of course world of parallel distributed file systems is very |
| 25 |
versatile, so for different tasks/workloads different file systems |
| 26 |
are the most suitable, but for typical IB-based HPC storage I see |
| 27 |
no better solution than Lustre at this moment. |
| 28 |
|
| 29 |
Best regards, |
| 30 |
Andrew Savchenko |
Archives