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,
two things:
This Ryzen offer is operating with local storage, NOT CEPH.
Its operating with ZFS Storage that works on 128k blocksize ( which is natural for ZFS ).
So any value below 128k will be unnaturally low.
About this whole yabs benchmarking with IO:
I suggest you to check how yabs is testing. They are using fio to test, and the settings they are using are optimized for good old local blockstorage.
That means that as soon as you are using some modern like filesystem ( CEPH, ZFS, ... what ever does not operate on the old 4k blocksize and will in general behave differently ) you can just not trust this numbers. What ever is written there, just forget it. Its if ever a hint to get a clue. But its just not reality.
If you are just using a disk ( no matter if spinning / ssd / NVMe / tomatoes / what ever ) and using a vanilla FS like xfs, ext3/4 or LVM, THEN this yabs alias fio will show something that is close to reality.
But if not, then simply not. And that does not mean that the servers are bad or slow or what ever.
Or do you really think that lokal NVMe drives in a ZFS can perform with 4 MB/s write or read speed? Or even CEPH or even a USB drive. No, it cant. Should be quiet obvious actually
This yabs fio testing can be simply ignored as soon as you just see complete unrealistic numbers.
That goes for both directions. So if you see that what ever works with tons of GB(it)/s then you will know that this can not be reality aswell.
There are other tests like https://bench.sh/ which are not using fio but dd. Its testing with 512k bs which is also not realistic, but will be in more situations more accurate, but still no reality.
IF IO performance is that important for you then you will just have to ask the provide what they use. Then you will know what you can expect. And if you dont have enough knowledge about that, then you will have to build up some minimum knowledge and understanding since you are about to rent a server as a serveradmin and maybe need some base for judgement to be able to compare offers between different providers.
in the end gb6 score is interesting. ideally 3000+
Hi,
yes, this geekbench is a very clear thing that only depends on the CPU capability to math what ever algorithm(s) this geekbench actually maths. Its perfectly compareable and not depending on anything else but the test subject...
Also if RAM speed would be tested ( which would be actually with the cpu a quiet useful test ), it could give you a good information about what performance this system will be able to deliver.
That would be nice to build it in...
Sorry, I assumed that you use Ceph because you say in your knowledgebase:
I think that the geekbench score does (a bit at least) depend also on memory speed, if not it could simply be tested with dd+ramdisk.
It might come down to the NVMe models used.
Using consumer-grade drives without power loss protection (PLP), or even worse, without DRAM, will result in a significant performance hit.