June 6, 20179 yr 1 hour ago, johnnie.black said: Never came across such bottleneck, but not quite sure I follow what's happening when you're hitting it, can you better explain what you're doing on the 1st screenshot? Thanks, what I doing is reading data disks in same time, perform file hash check locally. Before upgrade CPU, the max hash speed was 500MB/s, so I won't reach the limit. But now I upgrade to Ryzen 1700, I haven't hash speed limiit currently, i.e. total HASH speed could be, Array (700MB/s) + Cache (220MB/s) + SSD UD (400MB/s). I search this forum, also found some people got similar issue. Edited June 6, 20179 yr by Benson
June 6, 20179 yr 19 minutes ago, Benson said: Thanks, what I doing is reading data disks in same time, perform file hash check locally. Before upgrade CPU, the max hash speed was 500MB/s, so I won't reach the limit. But now I upgrade to Ryzen 1700, I haven't hash speed limiit currently, i.e. total HASH speed could be, Array (700MB/s) + Cache (220MB/s) + SSD UD (400MB/s). I search this forum, also found some people got similar issue. Really doubt that's an unRAID bottleneck, most likely it's your server, but I'll confirm later on one of my servers.
June 6, 20179 yr 3 hours ago, Benson said: There are bottleneck on R/W, the disk array will ceiling at ~700MB/s, those can't overcome by reconstructive write mode. Anyway, today just try 6.4rc2. The 1st image show still have bottleneck, a parity protected array will max at 733MB, 4 data disks already reach the max, last disk got ~4.5MB/s). These kind of bottlenecks are typically encountered if your SATA controllers only use 2 pci lanes (pci 2.0 ) or sometimes if cpu (old cpu's) speed can't keep up or if dma writes to ram are bottlenecked. If this is not the case in your system, maybe unraid's buffering is not optimal.... needs investigation. 3 hours ago, Benson said: With this limitation, the array write speed will also affect, it worse if more more data disks. The 2nd image just show all disk / system could run in their max speed. ( 1.6GB/s + 224MB/s, parity check+cache disk read) How did you accomplish this? Remember that read speed is one thing, but write speed even with recontructive write can only be as fast as your slowest drive
June 6, 20179 yr 2 hours ago, Videodr0me said: How did you accomplish this? Remember that read speed is one thing, but write speed even with recontructive write can only be as fast as your slowest drive The write speed would be the slowest read disk was true if no read bottoleneck, pls think if 700MB/s was max, then write speed for 2 parity 6 data disks system would be 700/5=140MB, for 7 data disks would be 700/6=117MB ..... so with increase in data disk, you should found R/M/W would faster then reconstruct mode. Edited June 6, 20179 yr by Benson
June 6, 20179 yr 7 minutes ago, Benson said: The write speed would be the slowest read disk was true if no read bottoleneck, pls think if 700MB/s was max, then write speed for 2 parity 6 data disks system would be 700/7=100MB, for 7 data disks would be 700/8 ..... so with increase in data disk, you should found R/M/W would faster then reconstruct mode. Yes, this is true if you have a hard limit of 700mb/s for all drives (maybe due to pci lanes, cpu speed or other reasons). Thats why i am surprised you got 1.6 GB/s+ read performance when reading concurrently all disks.For example if pci lane saturation would be your problem, i would have expected that you cannot not exceed 700mb/s in any case..... This has been a problem for a number of users. Maybe you could provide some more info about your system (cpu, mainboard, pci controller, sata controller) and how you got the above numbers (tunables?, direct io?)... Edited June 6, 20179 yr by Videodr0me
June 6, 20179 yr 10 minutes ago, Benson said: The write speed would be the slowest read disk was true if no read bottoleneck, pls think if 700MB/s was max, then write speed for 2 parity 6 data disks system would be 700/5=140MB, for 7 data disks would be 700/6=117MB ..... so with increase in data disk, you should found R/M/W would faster then reconstruct mode. This is incorrect, I can write with turbo write on at 150MB/s sustained to a 30 disk array, combined read speed is >4GB/s. Edited June 6, 20179 yr by johnnie.black
June 6, 20179 yr Quote This has been a problem for a number of users. Maybe you could provide some more info about your system (cpu, mainboard, pci controller, sata controller) and how you got the above numbers (tunables?, direct io?)... The controller was LSI 2308 chip HBA card ( i.e. 9207-8i ), it direct to CPU's PCIe. The tuneables and direct I/O also can't help. Edited June 6, 20179 yr by Benson
June 6, 20179 yr Just updated and VNC remote seems broken. My VM starts up fine but VNC remote won't connect.
June 6, 20179 yr 3 minutes ago, Woodpusherghd said: Just updated and VNC remote seems broken. My VM starts up fine but VNC remote won't connect. Read here.
June 6, 20179 yr 5 minutes ago, dlandon said: Read here. Read there. No solution. Any suggestions? Bug?
June 6, 20179 yr The noVNC client in unRAID 6.4 will now proxy the websocket request through nginx (Eric explained this to me). This requires that name resolution works properly. For example if you access the GUI using //tower then the browser must know how to resolve the name "tower" into the correct IP address. You can test your VNC by entering the GUI using the IP address of unRAID.
June 6, 20179 yr 4 hours ago, johnnie.black said: This is incorrect, I can write with turbo write on at 150MB/s sustained to a 30 disk array, combined read speed is >4GB/s. If combined read speed > 4GB/s , assume with 2 parity, then 4GB / 27 = 148MB/s, so it should valid. For my 700MB issue, I think it may relate to BTRFS, anyway no solution yet. Edited June 6, 20179 yr by Benson
June 6, 20179 yr 16 minutes ago, bonienl said: The noVNC client in unRAID 6.4 will now proxy the websocket request through nginx (Eric explained this to me). This requires that name resolution works properly. For example if you access the GUI using //tower then the browser must know how to resolve the name "tower" into the correct IP address. You can test your VNC by entering the GUI using the IP address of unRAID. I constantly access my unraid box using the IP Address and never use the server name. still wont work for me .
June 6, 20179 yr 28 minutes ago, bonienl said: For example if you access the GUI using //tower then the browser must know how to resolve the name "tower" into the correct IP address. You can test your VNC by entering the GUI using the IP address of unRAID. That does the trick. Thanks.
June 6, 20179 yr 51 minutes ago, Woodpusherghd said: That does the trick. Thanks. Doesn't seem to work for me at all, any recommendations/advice. screen shot added...
June 6, 20179 yr 3 hours ago, Benson said: The controller was LSI 2308 chip HBA card ( i.e. 9207-8i ), it direct to CPU's PCIe. The tuneables and direct I/O also can't help. Strange that a PCIe 3.0 x 8 controller should have any problems at all. Maybe you should open a defect report with all diagnostics. Maybe unraid/slackware does not configure that controller properly or its an unraid anomaly which comes to light on your system. Edited June 6, 20179 yr by Videodr0me
June 6, 20179 yr 8 hours ago, sarf said: Doesn't seem to work for me at all, any recommendations/advice. screen shot added... Same problem. Tried Server name, ip, http, https etc... no change.
June 7, 20179 yr 9 hours ago, captain134 said: Same problem. Tried Server name, ip, http, https etc... no change. I don't think you are accessing the UI on the http or https://tower links as in my tests it redirects correctly. Take a look at my last post on the following thread, as I give examples of how accessing the UI over http or https with and without the hostname gave me different results. Also if you take the URL that is failing from your address bar in the screen shot and replace the IP with tower VNC should connect as long as tower resolves on your network.
June 7, 20179 yr Just added a host name to IP address on my ubuntu box for my unraid server and it does indeed now work using the dns name! However, I have a lot of machines on my internal home network and no internal dns server as I simply do not need internal dns at home. For me personally I would 100% prefer it work with IP address first and foremost with internal dns as a nice to have but certainly not necessary to use web vnc functionality, Thanks for your reply and info Tuft
June 7, 20179 yr @sarf so you get the same results from http and https? I did not test prior to updating my hostname but afterwards I did notice that HTTPS was still working with WebVNC even though it did not show tower in the URL, maybe it was using it under the hood I've not verified. Generally most home routers have the option to give a DNS name to systems locally on the lan. Do you use a router of some sort as your DNS or use Public DNS servers? I've not tried this, but could register a FQDN pointing at your local LAN address. I could setup a subdomain for you to test with if you tell me the LAN IP of your server.
June 7, 20179 yr I just ran a few test's with the following results. All from my ubuntu workstation that now has name to IP entry for my local Unraid server. ------------------------ http://(internal ip)/dashboard [click vm link fails] http://(dns name)/dashboard [click vm link works] ------------------------ ------------------------ https://(internal ip)/dashboard [click vm link works] https://(dns name)/dashboard [click vm link fails] ----------------------- Hopefully there will be a fix at some point so that the webvnc works in all scenarios. I can set up a dns server at home no problem just prefer not to. it's a bit overkill for what I do and I remember ip addresses better than dns names any way but thanks for the offer
June 7, 20179 yr I have been getting the following error message: /bin/sh: line 1: 12686 Bus error /usr/local/emhttp/plugins/dynamix/scripts/monitor &> /dev/null Any thoughts and should I be concerned?
June 7, 20179 yr @sarf It's scary how easy it can be to remember IP addresses, I have a habit of walking into customer sites and can remember all the IP addresses for the machines I've deployed. Generally from 2 to 20+ machine deployments with Test platforms as well.
June 7, 20179 yr [mention=69318]sarf[/mention] It's scary how easy it can be to remember IP addresses, I have a habit of walking into customer sites and can remember all the IP addresses for the machines I've deployed. Generally from 2 to 20+ machine deployments with Test platforms as well. I forget how old I am.... never mind remembering IP addresses Sent from my LG-H815 using Tapatalk
June 7, 20179 yr There are some well deserved kudos for the new goodies in the 6.4 version: - The Azure and Gray themes are awesome. My personal favorite is Azure. - The responsiveness of the webUI is better than I've ever seen. - The inclusion of IPV6 and the added networking features is excellent. Being able to assign IP addresses to a docker is much cleaner in my mind than port management for one IP address. Many thanks to the LT team, bonienl, and all others involved in getting this version ready. Job well done!
Archived
This topic is now archived and is closed to further replies.