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.
VPS providers playing games with `dd` I/O results?
Hi everyone,
It is sometimes stated that certain VPS providers tweak or optimize their systems to provide an unrealistically fast result for dd disk speed tests. The implication is that the VPS providers with the fastest dd results may not necessarily give the fastest real-world I/O speed. Is there any truth to this? If so, what exactly do they play with to make dd return inflated speed results?
Thanks for the replies.
Comments
My cloudatcost VPS is stability at 6MB/s
DD in general does not represent real world I/O so using it as a benchmark is flawed to begin with. I do know in the past some providers have done things to improve DD results without actually improving IO. Run a real IO benchmark to get real results.
Like? Is ioping suitable?
Add it to popular VPS host rumors
ioping gives a good idea of IOPS but there are more suitable disk benchmarking software out there if it's that big of a concern. DD is good for a quick test but you shouldn't be looking at how big the results are and instead making sure the results aren't insanely low (like 0-10MB/s, but even then it's not an issue for something like a VPN that rarely uses IO),
we to used such thing, in simple "ploop" . but you can find it easily if overloaded,if their swap usage is large and wa on top
I agree overselling wont make sustainable business
I only use DD to do quick tests to see if there's anything abnormal on my servers.
It would be cool if some changed dd to a bash script that pipes the command to an actual dd binary, then replaces the speed in the output with a random very high number
This was discussed a long time ago on here. Basically somebody was going to change the DD script to just "echo" the output of their choosing and replace dd in all of the OS templates. That would probably be a quick trip to the deadpool though.
I know of at least one provider who optimized their RAID setup so they could get the best DD tests but it hurt regular IO performance under normal usage, this was during the "dark ages" when DD tests were actually used to determine if a provider was good or not (crazy right?).
Why bother optimizing? If I was a provider, I'd compile my own dd that gives the results I want and put it in the OvZ templates.
You would need to pin coreutils or dpkg-divert dd though
Would be quite obvious to an experienced user that something is wrong.
True, but I'm thinking that even aliasing in /etc/bashrc or something would fool most.
An experienced user does not rely on dd to decide if a provider is good or not. That "test" if for the others.
I've actually learned not to do DD by now, and only send that to providers as an indication of how shitty their write speeds are. From running the first dist-upgrade, I can already tell if there are any problems :P. Its quite evident if unpacking the kernel takes ages.