New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Good chance
When I posted it here we just kicked it off in the WHMCS queue. New WHMCS sends 50 emails every few minutes. I've scripted it to hopefully go faster and can confirm its hitting inboxes.
Glad you're enjoying it.
I intend to completely rewrite the redis handler. At the moment we have it wired through
systemd, but its systemDeezNutz sometimes. I think moving it into our own process handler would be more sane, and would mean we can catch when to auto restart.Francisco
NameCrane moved LUX to The Netherlands at the start of June. Completely live, maybe ~10 seconds of downtime while it did final RAM syncs. Our tech is going to be onsite in the next hour or so to rack up at least half the new nodes.
We'll kick off live migrations which will take ~6 hours to complete.
Francisco
I liked Luxembourg; it was special; it felt exotic. Netherlands is so... common.
Yeah, I know, I liked it too. It became untenable and with BuyVM pulling out, it was too difficult for Crane to justify being there. For shared it doesn't really matter as much, "exotic chickens" isn't really a thing like it is on the VPS side.
Francisco
There's no price/feature changes to active users or any sort of sunsetting planned. Any changes will come in the form of new plans and people can shift over if they'd like.
Everyones getting upgraded, we're not leaving any nodes on Ryzen's. It now means I have a massive pile of Ryzen's all over the world that I have to find a use for (or sell I guess?).
Singapore has cost Mike & I many nights of sleep. It and a couple of the NL nodes get rampaged by AI crap. SG is currently in "shared colocation" since it was mostly testing the waters. Its not possible to put the Genoa's in there though, they're too power hungry.
Getting our own rack is the only way of getting things upgraded, so that's what we're doing.
Francisco
For anyone thinking about using using it, I wouldn't if your application stops working when redis is down. I've had numerous downtimes caused by this issue when Namecrane restarted servers and redis wouldn't restart. I reported it on 31/01/2025 and it was still happening as of 11/05/2025 (with no answer on the ticket for months). I've since moved away from redis given it's instability.
Hey Fran -
So if nodes are moved from Ryzen to EPYC which has a bit lower boost clock(from 5GHz advertised to 3.7GHz), don't you think there would be a performance hit? Especially since most of starter web hosting packages are singlecored.
The current redis system is a bit derpy and auto restarts don't work at all. If you change a users plan in any way (resources, suspend, unsuspend, etc) then redis is killed every time, and it never restarts.
Handling the processes on our own is the way to go and its something I'll be coding one of these days.
The Ryzen's better if the node is empty/quiet and you can hit a full thread all the time. Problem is, thats rarely the case in shared/reseller hosting so you're in contention with other users, system tasks, backups, etc.
When backups run or things get busy we have Ryzen's that have < 5% idle CPU. Everyone on there, including system tasks, are all getting capped down to whatever cycles it can get (read: jack diddly squat).
The genoa's give so much breathing room for everything to run w/o a care in the world. We have so much spare CPU we gave Jetbackup 10 cores for shits/giggles, that's why its completing in 1 1/2 hours, and not 6h~18h.
Francisco
okay. except this "redis" the servers are good. i hope it can get auto-started in coming future.
i don't think that would be a performance hit when loading a wordpress page as many people use caching these days. But, the php is singlecored and to build a page it may take some mini-seconds more (its just a guess as of now) if we compare 4+ Ghz speed with the new 2.5+Ghz speed. But again, it seem not viewable to eyes while browsing. May be it can affect a bit to internal cache preloading or querring a huge size database and then building a page. Still my node is not epyc but within a week I will check the performance.
hope we crane luvers get some extra love from genoa's in terms of resources.
Congratulations @Francisco
Always great to see providers like Francisco who truly care about their clients and keep an eye on both hardware and software performance. That AMD EPYC 9654 setup with 96 cores, 192 threads, and 384GB RAM is absolute monster territory.
I appreciate that our hosting industry is moving towards better hardware; we’re no longer stuck with E3, E5, or underpowered Ryzen setups. After RackNerd's recent upgrade, it’s awesome to see you stepping up even further.
Honestly, your hardware is well ahead of the competition. Hats off, Francisco.
These CPU's run more like 3.5~3.7Ghz. Again, in our case, everyone wanting CPU is going to eat into that '5Ghz'. There's only so much CPU to go around.
Francisco
wow that's huge, namecrane++
Will there be IP changes upon completed migration in Netherlands?
@Francisco Can you tell me the exact specifications of the AMD Ryzen 7900 server? I mean, what motherboard, RAM, and hard drive model? I'm curious if you've had any problems with this processor, which is typically a consumer processor.
Nope. We own all of our IP's. For NL there'll be maybe a couple minutes of downtime to apply new RAM/CPU limits, but past that itll be painless.
Even SG/NY/UK shouldn't require more than that reboot to finish as we'll keep the Ryzen's running and just live migrate over the wire.
Francisco
B650D4U.
Servers are great, works well. It just doesn't have enough CPU threads for the work loads they're dealing with.
Francisco
Great news! Thanks Fran, this is an amazing upgrade. I’m confident it will solve the random timeouts for things that need to connect to the database.
Hmm, interesting… I’ll slide into your DMs.
What cooler did you use? They are in 1U chassis? On the Frantech discord, several people wrote that these Asrock boards can burn CPUs and M2 drives.
We had a few x470's go nuclear and fry some PCI devices. The B650's have been good for me but there was a batch of bad boards that just randomly died for many people.
Francisco
Shared web hosting is so funny. Hardware cycles repeat. I remember moving to the high-frequency E3s, which crushed everything in single-threaded performance. The problem was stability and handling spikes, especially as they started showing age. It eventually led to moving to slower Dual E5-2600 series CPUs with higher core counts, which improved stability and, for the average user, provided a much better experience.
Ryzens could run circles around our Dual E5-2600s, but at that point, we had just moved to the high-core-count EPYC/Xeon Gold CPUs. Sure, if the server is really underloaded, a user could get faster loading times on that 5Ghz Ryzen. However, you couldn't match the stability of a 40+ core system when it's fully loaded. We've been using the 9654s lately, and at 3.7GHz, they pack a punch. The 96 cores are a lot, but you can always split them into two good-sized 48-core shared hosting servers, providing a way better experience than most.
I'm sure some users will be unhappy with the change, but overall, most users will be happy with the consistent performance, as opposed to the performance peaks and valleys that were undoubtedly happening before.
Very much what we've seen
All but 1 node has been swapped over to the Genoa's in NL. That one will be completed next week.
Feedback is always welcome!
Francisco
Can't wait for the update. Since the migration to The Netherlands I saw the load got double most of the time but still everything works great and fast till now. I am to curious how it's gonna be with the new upgrade 🤔.
>
Only DA03 is left to do, everything else has been migrated
Francisco
@Francisco what happened to the plans to offer shared/crates in Switzerland?
I'm not in a rush to do it, we have all of this to sort, domains, etc.
Francisco
Agreed.
We've basically never had to worry about CPU utilization on our EPYC 96 or 128 core hypervisors, compared to our Ryzen's where it can be a constant worry. Performance is so much more consistent and stable on EPYC, we've basically never had to worry about them. There are just too little cores on the Ryzen's.
I expect NameCrane clients to have a much better experience with the EPYCs than the Ryzen, even if on paper, the clock speed is lower. Zen4 Genoa still offers great single threaded performance while offering a boat load of cores.
Seems like my package got migrated, can confirm major improvement
@Francisco I can also confim improvment in loading wp-admin pages (non-cached) too fast in comparison to ryzen. system load is 5. cached ram is 243 GB.
btw, can you tell what data this ram caches? It seems opcache/apcu data and what else? im wondering.