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
Hi,
depending from where you come geographically the latency will be a bit better or not.
The major difference between the locations is:
FR = Xeon SP 6138 while DE = Xeon SP 6122 ( thats 0,2 GHz difference )
In fact the rules to be entitled to get one of the servers was too easy and this way we got far more orders than expected.
So i had to update the rules. Especially the 5 TB DE server is now much more tight to get. And the number was limited now to maximum 5 additional accounts.
We are still working up the orders and will prioritize those who managed to give their LET username in the order, like requested, so we dont need to run behind this information and open tickets to ask for it.
The 5 TB DE will run out soon i assume. 10 TB DE we will keep up for the moment without limitation in numbers same with FR.
Hi,
if you dont take the smallest possible configuration, there is always space for negotiation. Simply contact us.
But in general there wont be that heavy discount like we have it now for the storage servers.
Aside of that, there might come another event like this in the coming months that will focus on compute ( CPU / RAM ).
Good
I will wait it
I hope I'm not the first one to try this.
Hi,
for sure not, and if yes, well welcome
To give this benchmark more sense:
Edit the yabs testing file, there is an array ( line 610? ) where you define the tests ( "4k" "64k" ... ).
Add the 128k to match your FS blocksize.
At latest then you should get the full 80 MB/s while your paste looks a bit strange.
There should be a limit at 500 IOPS but in 4k and 64k you get significantly more. Looks strange. Are you sure you execute this on this storage servers? ^^;
default yabs.sh FIO shows up for me as
Hi,
yes, looks basically correct / realistic. Either you hit the 500 IOPS cap or the 80 MB/s cap ( give or take some % ).
wget in 5T disk writes at ~66MB/s and on the primary disk writes at 53.8MB/s
Overall - works well as a backup node. I wouldn't go installing a large nextcloud instance!
5T:
wget on "primary disk" >!
@layer7 cpu performance sadly is deteriorating very quickly, at least in germany
Geekbench 6 Benchmark Test: --------------------------------- Test | Value | Single Core | 459 Multi Core | 813 Full Test | https://browser.geekbench.com/v6/cpu/12222874I assume people start putting all kind of crap on it (windows, scrapers, torrent)... any additional balancing possible? like putting the high CPU abuserall together on a node 😄
@layer7 guess DE is OOS... sigh...
Hi,
please open a ticket. From what i can see, hostmachine wise the overall CPU usage is around 30% ... so maybe worth to take a look at specifically yours.
yes for the 5 TB version.
10 TB is available. On the FR location both are still available.
Just FYI: I'm seeing the same (around 25% slower than yesterday, also in Frankfurt) with still basically no steal, but maybe that's just because it can't boost as much anymore.
Hi,
theoretically, if you have no steal, and you can actually have no steal as the HostCPU is fine, CPU based benchmarks should not show something else like before.
Or you do not benchmark with 100% idle system ( while i assume you wont do such mistakes ).
Or maybe the upcoming IO waiting time ( which should, if you idle the system have no impact ) is concerning the benchmark software somehow.
Or, and thats most probably what we see here, the benchmarks before during system idle where better as the hostmachine could perform more operations in the turbo mode than it can now when the workload is stacking up.
So maybe its just not getting weaker compared to before... maybe it was before just better than it would normally be...
Please let me know via ticket if you will have some real life problems like your application wont work anymore while it should or something like this.
Also in general currently there is a lot ongoing network wise. People seems to fill the cavern... lets see how things develop.
I'll give it a day or two. Currently system is idle/empty with debian 12 from template.
Single score 500 just seems quite low for xeon gold, but to be fair I didn't check deeper on the specific type what to expect.
Plan on running PBS and borg on it. The two cores probably will manage, but if they are more potent short time/burstable the jobs can easily benefit from it timewise 😉
Anyways, will contact you eventually via ticket then...