https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# longhorn-storage
a
This message was deleted.
p
Yes, that sounds about right on v1 engine.
t
i see the longhorn wiki benchmarks shows performance very close to local storage, but it has been tested on very slow disks. Is it perhaps just a limitation of longhorn that haven't been possible to test? We're getting a node with PCIe5.0 disks soon, if we see the same ~1GBps read on those then it has to be a problem with longhorn? The specs of the node btw is two Epyc 9374F with just shy of 800 GB of RAM, filling up all memory channels.
p
Also strict local is slower than multireplica in sequential speeds, even in v1 (and in v2, multireplica can be faster that local if you have the network speed)
t
we only have one node with this spec at the moment. 100Gbps is on the horizon too. just waiting for our second node
p
But yeah, while longhorn is basically the fastest foss way to use replicated storage over multiple datacenters, its overall wayy slower than local storage. v2 engine is supposedly much better doe
t
the v1 vs v2 numbers i'm seeing on the wiki aren't that much different though
i'll put it on my todo list to try with it after hours
seems to really help RWX performance though, which we also require for other workloads
p
Sequentials theyre somewhat on par, but IOPS gains are absurd on v2 (v1 being super slow)
t
oh yeah, didn't read those
are there any downsides with v2?
p
Currently its not production ready
t
this is mosly dealing with build cache and such, so the data isn't super important
s
@powerful-librarian-10572 any idea when v2 might be considered production ready?
p
1st - Never ask for ETA, it's impolite 2nd - No i have no idea i'm just a regular user lmao
t
so it's still experimental in 1.8 from what i understand
p
Yep. Its a complete rewrite of the longhorn engine sooo lot of work ahead
t
general avaliability(?) in 1.9, https://github.com/longhorn/longhorn/issues/8689
wondering if we have to bite the bullet and get ourself a all flash truenas box or something of that sort...
p
If your volumes are not data-criticial, no harm of enabling the v2 engine
Worst case scenario is the volumes not working after an upgrade, requiring manual intervention. But if the data itself is not important, it might be worth a shot
t
yeah. currently we only have one node it can live on anyways, so it might as well be emptydir
p
One-node cluster is bad doe
t
we have 17 or so nodes
i think five longhorn nodes
p
oh sorry
t
we're slowly replacing our old servers
49 Views
Open in Slack
PreviousNext
Powered by