Discussion in 'General Software' started by elvis, Feb 10, 2021.
I thought you had a 10G network?
Yes I do, that’s why the speed seems terrible. I’ll have to move that cache server on to something else hardware wise. It must be running through on onboard NIC I’m guessing.
Is yours on metal or VM?
It's running on my Ubuntu box, with the cache on a NVMe drive to make good use of the 10G network.
Good-natured fun is all
Found the issue, it's hanging off a 1Gb switched link. I might have to get a little crs305.
I have heard anecdotally / second hand of LAN organisers who flood 10GbE connections with LanCache backed by a SSD/NVME ZFS setup.
Seems like the software itself can scale quite happily to whatever hardware you can throw at it. It's not super complex, just a couple of DNS hijacks pointing to a local storage device exposed by HTTP.
In my setup the bottleneck is the client machine, especially if it is saving via a SATA interface.
J2C had a bit on this for big LAN's around the 16:24 mark, it does scale up well -
I had to roll my eyes hard at the "people want to be on the last port of the switch" bit.
Gamers getting a hold of technology-old-wives-tales is the worst.
Yeah, I had a giggle about that bit as well. I'm guessing they also used that port for the uplink to limit the risk of that happening, as I'm guessing there would be some who would want to be in port 1 for much the same reason.
Maybe I've been an enterprise/cluster/renderfarm/HPC sysadmin for too long, but the whole video made me laugh in places. Very amateur stuff.
This is why we use bungee cables from above per desk, and don't let numtpies anywhere near the edge switches that are locked in a cabinet. And all the brand name bragging was chuckle-worthy too. Cisco this, Juniper that. For gaming over gigabit Ethernet, it makes three fifths of fuck-all difference.
People really need to worry less about theoretical nanosecond delays and get on with playing their games.
Anyways, what *was* interesting was the 50Gbit/s+ they were able to push out of their cache. Again, I'd dare say you could just keep going as long as you had the disk and network interfaces to enable it, especially at a LAN where you could go quite parallel in your design.
Just set up lancache at my girlfriends house - Core2Duo E7500 with 4GB ram and a 7tb spinning rust disk, no problems with 4 gaming PC's and 400-500 Mb/s transfers on a gigabit network. Much better than waiting for everyone to download at 40Mb/s off the NBN.
I had forgotten about this; Justin said he updated the container since they changed the upstream dockerfile.
The Lancache on RPi Project; https://github.com/jrcichra/lancache-rpi
You know as well as I do its spares from work...
Haha it certainly was for a decade or so there.
Today, it's AliExpress 2.5GbE goodness.
I've noticed lately that while subsequent downloads come off the cache at good speeds, initial downloads appear to be hobbled. Downloading from Blizzard on the PC or from Microsoft on the Xbox seems to sit at around 50Mbps yet if I bypass the cache I get at or close to the full 100Mbps I have available.
The Raspberry Pi it's running on doesn't seem to be maxing out its CPU and the USB connection to the mechanical HDD should be able to talk in 100Mbps worth of traffic in/out concurrently....
Has anyone experienced this?
Yes. I get very poor download speeds with anything Microsoft, 50Mbps on a NBN 1Gb (ha, more like 400Mb) link.
Steam and Epic seem okay.
Download speeds also tank on everything if another client starts hitting the cache, but that's most likely an IO issue with my dodgy setup.
I recall reading in the lancache documentation somewhere that using the USB storage can be a bit flakey. May or may not be relevant.
What model RPi?
Full path will be:
* Network in
* Write to spindle
* Read from spindle
* Network out
So the hit is 2x the network and 2x the disk for that transfer. Is there something else you can verify in a similar fashion? Try doing SCP file copies to and from your Pi simultaneously and see what the performance is like?
I'd definitely want to see some benchmarks, but my gut feel is the Pi's specs could potentially be the issue (depending on model and other things).
I'm running off an old Intel i5 chip to a dedicated SSD just for LanCache stuff (server does heaps of other things too, but not from that drive), and zero problems.
Most "RPi USB storage is flaky" observations come from lack of power over USB, and the USB drive resetting. You can see these outputs in "dmesg" on the command line (look for messages about your SATA or ATA controller).
I have a couple of RPi4's out in the wild as backup devices in small businesses, and always use USB3 attached disks with their own power bricks. With that in place, zero issues as they're not drawing power over USB, and no issues.
Generally SSDs in USB caddies are OK as they're much lower amp draw, and only need 5V. 2.5" spindles are a bit heavier on juice requirements, and 3.5" spindles often need the 12V and mandatory external power.
As it happens so am I, it’s just a old Dell Micro.
There are pre-caching options for game files.
Have had this running on my Unraid box in Docker for a year or so now.
Not really a requirement as I am the only gamer but it does make re-installing Windows very quick these days.
Touch wood, have not had any major issues.