Tom3
-
Posts
104 -
Joined
-
Last visited
Content Type
Profiles
Forums
Downloads
Store
Gallery
Bug Reports
Documentation
Landing
Posts posted by Tom3
-
-
During my 6.12.8 upgrade, the interfaces did not do a link down. This resulted in
the DHCP server re-assigning new IP addresses to all the interfaces on UNRAID.
Once I found those, everything came up OK on the new addresses.
-- Tom
-
The update to Firefox 113.0 has resolved the noVNC problem. It appears that 112.x.x versions have the issue.
-- Tom
-
Speculation...
If sufficient errors were detected, the link might auto re-negotiate the link speed.
That might fail for 1 GbE, then retry and succeed for 100 MbE.
Can you disable auto negotiation, and set the link to only 1 GbE at the router? At the server?
-- Tom
-
Hi Mike - early on, the equipment on each end used Link Aggregation to package 4 10GE links as a single 40 GE link.
This is where some issues cropped up.
The ratified version of the 802.3 ba standard (June 2010) for 40GE & 100GE inserted something called
Multi Lane Distributor (MLD 64/66b) to handle all this in hardware chips on the Ethernet ASIC
interfaces ahead of the physical interfaces. (The physical interfaces generally don't know about this).
Your Mellanox interface is one end, the ethernet switch that the cable is plugged into is the other end.
So the question is what is the age of the equipment on each end of the link? If both ends are listed as
being 802.3 ba compliant, then you should be OK. Practically most equipment took several years after the
standard rataification to adopt everything.
-- Tom
-
40GE can be a bit of a rat's nest. The physical standards are primarily 4-lane interfaces (4 copper lanes, or 4 optical lanes).
That usually makes the transceivers a bit expensive.
Many of the initial deployments of 40GE used it as 4 independent 10GE links with a special break out cable arrangement.
This was done as a way to increase the density of 10GE interfaces on a piece of equipment, not as 40GE links.
For 40GE use those usually had link aggregation (x4) to treat the bundle as a single 40GE point-to-point. If your application
is as a single-40GE-link (which is what it sounds like), you need to be concerned with compatibility of the two
ends, both physically and with link aggregation compatibility. Expect to need a fair amount of configuration effort.
There were a number of link aggregation issues back in the 2010-2015 time frame where the two ends just would not
talk between vendors. Whether older equipment was eventually harmonized I don't know (and is probably vendor and
model number dependent).
-- Tom
-
1
-
-
Found the problem.
It appears the VNC VM GUI display does not play well with Firefox 112.0.1 on the Ubuntu client. The previous days version
of Firefox worked fine, so likely the latest update (I think this is the most current version). I cleared browser cache
on the Firefox client - that did not improve anything.
I launched Chrome 112.0.5615.121 on the Ubuntu client, and that works well with the VMs.
-- Tom
-
1
-
-
Hi JorgeB - thanks for the reply! Using the console to transfer large files between NVMe and HardDrives
is very fast in both directions, no errors.
-- Tom
-
This morning, all VM's (old, new, recently updated, not recently updated) suddenly are extremely slow. On 6.11.5.
The UNRAID GUI seems fine and snappy, it's just the VM's. Opening a console window on a VM that worked fine yesterday brings up
a sluggish console window - very very slow to resize. Running 'top' in that window VM shows nothing consuming large CPU resources or
memory for that VM, adequate free space and CPU. Five or Six Ubuntu 20.04 and 22.04 VMs tried, all slow as molasses.
Running Firefox in the VM is almost impossible, 10-12 seconds to respond to each mouse click, but it does function correctly eventually. All services run by VM are almost ground to a halt.
Only update this morning was I think unassigned devices plugin 2023.04.17
Rebooted the server, no improvement. Server memory, disk stats look fine. Network working correctly.
Docker images come up fine and the GUI is very responsive, so this seems isolated to just the VMs.
Diagnostics attached.
-- Tom
-
Dynamix ssd trim plugin - is it still needed?
On Unraid 6.11.5
I am getting a plug-in file install error, referencing: /boot/config/plugins-error/dynamix.ssd.trim.plg
The button to remove it does not work.
The file is in fact in the /boot/config/plugin-error/ directory. Is this plugin still needed, or has it been superseded by upgrades to UNRAID over time?
If it's no longer needed, can it just be manually deleted from the containing directory?
-- Tom
-
Having difficulty updating to Dynamix System Statistics 2023.02.05a
When clicking the UPDATE button nothing happens.
Under the tab Plugin File Install Errors one item is listed:
/boot/config/plugins-error/dynamix.ssd.trim.plg
It shows a status of ERROR. When clicking the Remove button for that, nothing happens.
Have since rebooted the server, the above does not change.
There are no obvious errors in the log.
The graphical stats page continues to work
On Unraid 6.11.5.
-- Tom
-
Are you accessing it via an IP address (e.g. 192.168.x.x) or via a domain name / suffix
(e.g. mytower.myhomenetwork). IP addresses can sometimes be re-assigned by your
router (or ISP modem/router) on reboot which perhaps might change the IP.
If you are using a domain name, do you have a DNS that knows how to redirect it internally?
Does that need to have it's IP updated?
If your server is using static IP than that shouldn't be an issue.
Are you using My Servers service? (if so, that's a bit outside my experience).
-- Tom
-
I use Acronis for backing up various machines to Unraid. No share is mounted on Windows.
In UNRAID, a new share is setup as: Public, Export=Yes, Use Cache pool = No My backup speed is not limited by the UNRAID hard drive but by Acronis, so for me there is no value in using the cache for this.
In Acronis, create a new backup, and select Backup Destination. That will be Network --> name of your Unraid server (perhaps TOWER) --> Name of the public shared folder above. Acronis should (sometimes quite slowly) populate the path as you select each folder down to that new share.
After the backup completes, you probably want to limit access to that newly created backup file.
In the UINRAID command prompt:
$ CD to the new backup file's containing directory
$ chmod 444 the_backup_file_you_just_created
$ chown root:root the_backup_file_you_just_created
That will make the backup file read-only, and only the root login in UNRAID an change that. Acronis creates a new file for each
differential backup. I prefer NOT to use incremental backup. The difference is that differential writes the backup between now
and when the original full backup was done. To restore you only need the full and the last differential file to be good. Each
differential file is larger as a result.
Incremental writes since the last incremental backup. That means to restore you have to go back to the full, then replay all
incremental backups. If one of those is damaged your restore stops at that point. But the incremental file is smaller as a result.
-- Tom
-
1
-
1
-
-
You might find this Docker folder organizer to be helpful, from Ibracorp.
https://docs.ibracorp.io/docker-folders/
-- Tom
-
There are certain ASCII characters in the filename that don't work with Rsync on Linux,
rsync fails to copy the file as a result.
My recollection is the vertical pipe character | is one of them, colon is another, perhaps some related to
Linux file redirection such as greater-than, less-than might be others. Some of these are OK in Windows.
-- Tom
-
Playing with several Ubuntu VM's - found that logging out the local user (while leaving the VM up and services running) cuts down the write rate dramatically.
-- Tom
-
The storage management section of the manual details how to handle cache replacement.
Recommended reading.
https://wiki.unraid.net/Manual/Storage_Management#Why_use_a_Pool.3F
-- Tom
-
Hi Orlando - it looks like the Mellanox MCX311A-XCAT has been reported by others to work
correctly in Unraid (based on google search):
I'm puzzled why ethtool on both systems reports the interface (if) as link down.
-- Tom
-
Hi Orlando - interesting. ethtool reports 10000baseKR (which is the single lane
backplane electrical interface version of 10GbE). The coding on that is the same as
10GBaseSR, so it's likely just an unusual way to report the interface.
You don't report which driver the 10G interface is using. (ethtool -i devname)
Not all NIC cards are supported in Linux, you need to check to see if the driver and card
you are using is supported by UNRAID.
With optical interfaces there are some compatibility issues that need to be taken account
of:
1. Both ends need to be compatible (same wavelength, same fiber type, correct fiber type).
If you are using 10GBaseSR (short reach) multimode fiber (MMF) then the fiber needs to be the
right type. 10GBaseSR is not too stressful on the fiber characteristics, I think up to about
500m is normally OK. The two optical modules need to be compatible.2. That interface can support FEC (forward error correction). FEC needs to be set the
same (On / Off) at both ends. Depending on the software, it may need to be manually
configured. Your card may or may not support FEC - ethtool is showing 'Not reported'.
Perhaps you may need to set if to Off at both ends.
3. Fiber connector cleanliness is critical. Dirty connectors are far and away the most
common problem with fiber and can cause all kinds of strange symptoms. Both the fiber
end-face and the barrels need to be clean and unscratched..
4. ethtool message on eth2 reports that the interface is down, but ip show reports it up.
-- Tom
-
1
-
-
One place to start is using 'ethtool' from the UNRAID cli. It will
tell you about negotiated speed, duplex, other parameters. This
may tell you if something is misconfigured.
# ip link show
will list all the interfaces on your system (devname is usually eth0 eth1, eth2, etc.)
# ethtool devname
will list the parameters for that specific interface
#ethtool -i devname
will tell you what driver it's using
# ethtool -h
will give you brief help. Suggest reading up on ethtool online.
https://linuxhint.com/ethtool_commands_examples/
-- Tom
-
1
-
-
It probably depends on what you want to do with your system.
A critical element being whether you want to host dockers or VMs.
In my system I have 4 x HD and 1 x NVME SSD. Initially all the docker
containers and VMs were on SSD.
However one of the VM's (an Ubuntu server instance) was totally clobbering the
SSD with about 2 GB/s write rate. After one week it had consumed about 10% of the
SSD rated lifetime. There was a lengthy thread about Docker containers having write
amplification problems, and eventually a fix seemed to come about. But my server
VM did not appear to benefit. So I moved all the always-on VMs over to Hard Drive.
The write rate there is *****WAY***** lower than to SSD for some reason.
-- Tom
-
I'm not an expert in reading diagnoistics, but it appears that your system has the br0 and br1 NIC interfaces
bonded together into a Link Aggregation Group (LAG). Does the thing they connect to (Ethernet switch?)
support LAG and has that switch been configured for LAG?
You might want to try looking at your settings, and if bonded, eliminating the LAG bonding, then connect
just one of the interfaces to the Ethernet switch. The <your-servers-ip> /Dashboard/Settings/NetworkSettings page
is where you can re-configure the network settings.
-- Tom
-
The retr column shows the number of retransmitted TCP packets. The first row
in your transfer shows 56 retries, which is so high as to appear essentially non-functional.
So the question: what is the cause ?
Some things to check:
1. Ethernet cables.
2. Any intervening Ethernet switch.
3. Do you have two things assigning DHCP addresses on the same LAN segment?
4. Bad NIC card or connector.
After that it gets more difficult to troubleshoot:
5. Wrong driver for the Ethernet NIC.?
6. Out of memory condition preventing TCP from acquiring buffer space?
-- Tom
-
It's difficult to see what market this is targeted at. Large companies probably would use
some sort of tape mechanism that can robotically insert, remove and change tapes. Rolling
backups come to mind.
For hobbyist and home use the price is too high compared to just buying some 10+ Tb disk drives.
The 100 year lifetime also means that the drive and writing system would need to be around in 100 years.
So does that really means that the probability of failure to read is lowered in the 1-10 year timeframe ?
Perhaps the commercial photography and video market might be the target. Hence BH Photo selling it.
-- Tom
-
Too bad that the ODS-380U drive it goes into costs $8995.00
https://www.bhphotovideo.com/c/product/1549175-REG/sony_odsd380u_optical_disc_archive_gen3.html
-
1
-
[Plugin] Docker Patch 6.12.8
in Plugin Support
Posted
Thank you! Thank you! This fixed a tailscale problem.
Prior to the patch, tailscale was removed on update due to the error.
Fortunately Unraid keeps the previous template, so reinstall worked easily.
-- Tom