Vexorg Posted July 20, 2018 Share Posted July 20, 2018 Hello all, I've just built my unraid server and I'm in the process of moving content from my old storage systems to it. I have a two systems I'm transferring to to unraid over a a bonded connection using 802.3ad via a cisco switch that can also do it. I've configured the two machines in a way that they each use one of the connections. In theory i'll get 2x1Gig... On the unraid server I'm using krusader to transfer from old server to unraid, skipping a middle man. Other is from a windows machine to the unraid server. When i start the transfers the old server transfers at about 450Mbit, being a decade old server I can live with that. From windows I get the full gig. After about 4 minuets both transfers tank big time and never seem to recover. See the attached pic, it shows the two ports along with the port channel combined speed. Not sure if I'm filling a samba buffer or an MD buffer or something else. In the long run really does not matter as after these big transfers the unraid server will have all the old data on it. PS. I also turned on fast writes before this all started. Link to comment
ken-ji Posted July 20, 2018 Share Posted July 20, 2018 How much data are you transferring versus unRAID system RAM? You might be running out of system RAM used for caching after say 5m... particularly with the Windows box now pushing a 1Gbs or ~100MB/s + the first transaction at 450Mbps or ~45MB/s Link to comment
Vexorg Posted July 20, 2018 Author Share Posted July 20, 2018 I have 128G ram and the transfers are 600GB and 1.6TB. My drives are relativity fast with ~200MB/sec transfer. Also have cache drives but not using them as they are 500GB and seem pointless for these big transfers. Link to comment
ken-ji Posted July 20, 2018 Share Posted July 20, 2018 Hmm. Are you copying small or large files to array in turbo mode? because I'm fairly sure once the RAM cache is full, you'll start to experience slower disk writes (turbo mode still goes faster) specially with two transfers to the array. because now you have contention due to the disks needing to jump around to write to two different areas at the same time. unfortunately slowdowns like this are extremely noticeable when writing to the drive. Its going to be like trying to do a drive reconstruction and then running the mover to move a whole bunch of directories... Link to comment
pwm Posted July 20, 2018 Share Posted July 20, 2018 You should only perform one transfer at a time, since you get contention for the parity drive even if the actual data from the multiple copies goes to different data disks. And seek time on a HDD is extremely costly, so it doesn't matter what raw transfer speed the parity drives can manage. Link to comment
Recommended Posts
Archived
This topic is now archived and is closed to further replies.