0

I have an interesting problem/situation I'm dealing with in Eclipse. I'm running an application who's processing lots of data, several tens of gigs worth.

I have more than enough RAM for how this application is supposed to operate and a very beefy CPU. My local disk is the first issue, while the application processes this data I run out of space on my local disk due to temp files. I solved this by moving my temp directory to my NAS which is mounted using iSCSI

-Djava.io.tmpdir=E:\tmp

Here's the actual question:

When I switched to my iSCSI drive I noticed more consistent memory consumption and quicker execution by the application. Even with my iSCSI drive being in a RAID 10 over a link aggregated connection, I would actually assume memory consumption on the system would increase due to overhead and I would see a slow down in application execution, which isn't the case.

Is a reduced memory footprint and quicker application execution in this situation expected? If so, why? If not, where might I begin to look for a reason why this has occurred?

user316114
  • 803
  • 7
  • 18
  • You mut have a bottle neck somewhere else, and the garbage collecter ca now keep up, attach a profiler to find out where. Also, voting to close as off topic as there is no real problem presented. – Matt Clark Apr 03 '17 at 04:32
  • @MattClark I wouldn't say there's "no real problem", but this doesn't seem reproducible. – chrylis -cautiouslyoptimistic- Apr 03 '17 at 04:37
  • @MattClark and chrylis I understand if this gets closed, it's an ambiguous issue on proprietary software, I'm just at a loss for where to get advice on this – user316114 Apr 03 '17 at 04:39
  • Try using something like [YourKit](https://www.yourkit.com/) to profile the code. You can include one of their native libraries which inspects the bytecode at runtime. Using this, you can figure out where the hotspots / slowdowns may be. I mean, I would not be one to complaining about quicker execution and smaller memory footprint ;) – Matt Clark Apr 03 '17 at 04:41
  • And honestly, I would not expect much overhead at all. I would actually _expect_ the downsteam, link aggregated RAID 10 array to preform better then the local disks. With the iSCSI protocol as well, from the applications perspective, there is no overhead; the disk is looks as it would be as if it were local. – Matt Clark Apr 03 '17 at 04:44
  • Thanks for the suggestions, Matt. Funny, we have the same thought process. The reason I account for overhead is because the local disk is an NVMe SSD. So comparing link aggregated raid 10 to a local NVMe SSD, I would think there would be no debate that my NVMe drive would lend it's self to better performance over the raid. – user316114 Apr 03 '17 at 04:47

0 Answers0