TrueNAS Scale Released and Resetting the NAS Paradigm

[Patrick Kennedy]

TrueNAS Scale 22.02 hits release and it resets the NAS or hyper-converged paradigm in a big way with the new open source Linux solution

The post TrueNAS Scale Released and Resetting the NAS Paradigm appeared first on ServeTheHome.

Continue reading...

 

[tjk]

Love it, except for Gluster for scale out. Makes me sad.

 

[PigLover]

Didn't see anything on licensing model for TrueNAS scale.

 

[T_Minus]

oh, yaye, it must have been since I downloaded this Sunday
maybe I should order a GPU so prices go down

 

[tjk]

Didn't see anything on licensing model for TrueNAS scale.

Are they planning to sell just Scale software licenses? If so, that's news to me. I assumed there would be the FOSS version, and then the enterprise version (HA, etc) that came on their own hardware like Enterprise does today.

 

[Patrick]

@PigLover - nothing right now. I think they are going to build HA and other features on Scale and license those more like TrueNAS Enterprise.

I fully expect in 2 years one will be using Linux Scale instead of FreeBSD Core for even single box storage.

 

[gigatexal]

I fully expect in 2 years one will be using Linux Scale instead of FreeBSD Core for even single box storage.

Given ZFS on Linux development to be consistent amongst all the various platforms this makes sense.

 

[Rand__]

I think the main issue will be performance.
Its still to fresh to see good comparisons, but in a few days we will see how a single TNS node works out compared to a TNC node.
If that scales up in an acceptable manner using gluster for multiple nodes remains to be seen... they are good but not sure they can do magic

 

[tjk]

I think the main issue will be performance.
Its still to fresh to see good comparisons, but in a few days we will see how a single TNS node works out compared to a TNC node.
If that scales up in an acceptable manner using gluster for multiple nodes remains to be seen... they are good but not sure they can do magic

So far, lots of folks on reddit are saying the performance on TNS isn't matching the performance they had/have on TNC.

Tom @ LTS tested a beta release and also had performance complaints.

 

[Rand__]

Yes I've seen those, I hope its just that its not fully tuned yet.
In the long run I assume it will be en par for single node performance or even ahead, but am also afraid that its distributed side will not be able to keep up.
Network latency + extra layers will take their toll and it remains to be seen how good they can make up for that. Maybe roce/rdma will help there, but i think that still off a while (on official basis, even if it might be working already)

 

[tjk]

Yes I've seen those, I hope its just that its not fully tuned yet.
In the long run I assume it will be en par for single node performance or even ahead, but am also afraid that its distributed side will not be able to keep up.
Network latency + extra layers will take their toll and it remains to be seen how good they can make up for that. Maybe roce/rdma will help there, but i think that still off a while (on official basis, even if it might be working already)

Couldn't agree more in terms of single node performance and the future. I still scratch my head wondering why they took the glusterfs path for scale out.

Not sure how much life glusterfs has left tbh, I think RedHat is betting on a ceph/cephfs future and gluster will be put on life support, if it isn't already.

 

[Rand__]

Never run a gluster config, but when I was playing around with ceph (3 or 4 node, totally unoptimized, but all nvme) it was competing against a snail and lost...
O/c that was years ago, so maybe they improved some 10fold+ but if not ...
I dont think small clusters are what Ceph is aiming at, it gets its performance from large/massive amounts of distributed copies and i am not sure that Scale is really going for that.

 

[tjk]

Never run a gluster config, but when I was playing around with ceph (3 or 4 node, totally unoptimized, but all nvme) it was competing against a snail and lost...
O/c that was years ago, so maybe they improved some 10fold+ but if not ...
I dont think small clusters are what Ceph is aiming at, it gets its performance from large/massive amounts of distributed copies and i am not sure that Scale is really going for that.

You are correct here. I think @Patrick is using Ceph on his proxmox clusters, be curious to hear about his config and feedback on performance/use case/etc.

I've run Ceph for FS services and it just works and performs fast enough as an NFS replacement on < 10 node cluster.

 

[Patrick]

That is the challenge with Ceph. Really 3 nodes is too small. It is meant for big scale-out. I think iXsystems saw much of its customer base with <8-12 boxes and realized it needed a different solution.

On the Scale v. Core performance, it is 100% tuning. My Proxmox VE ZFS NAS runs wire speed at 10GbE easily. Debian + ZoL is capable of performance, but I think TNS is behind on tuning.

Once Scale gets going, the killer feature is really KVM + k8s in the same box. That lets TNS go after VMware's install base. TNC is stuck mostly just being a NAS. My sense, since Scale was launched, is that iXsystems has to invest in TNS and TNC will have to go more into a maintenance mode. The TAM for TNS is several times what it is for TNC.

 

[PigLover]

Ceph is best suited to give consistent performance at massive scale. It is not a high performance solution. Given a large enough cluster (likely minimum 10 nodes) and fast networking (25/100gig) it will meet the performance objectives of all but the most demanding DB workloads. But the real benefit is that same level of performance scaling near linearly to massive scale.

Ceph really isn't appropriate technology for small clusters, though it is also not horrible if its performance at that scale suits you application.

Not clear that Gluster will fare much better. My personal experience with Gluster has been quite bad.

 

[Sean Ho]

Ceph can certainly work on single node with OSD failure domain, behaving like an overachieving mdadm. But I think to improve small-cluster performance it'd benefit from RDMA and better use of memory caching. I'm happy with my little 5-node cluster, but I also don't expect single-node performance from it.

 

[trumee]

Dissappointed that they dont support LXD. Would have loved to run asterisk/freepbx in a container. Not everyone uses docker!