[opensuse] Re: [opensuse-factory] Phoronix has done some network tests windows vs a few distros (incl TW)
to discuss sloppiness of tests (or at least an unwillingness to clarify test assumptions) (its a while ago, but i believe this is the outline). in 2016 he did comparisons of different distros, testing read write speed using a database. After pointing out in the comments that in fact opensuse, btrfs guides and (i think) the authors of the database recommend using nodatacow for such applications the comment was 'set upon' by michaels fanboys using lame, handy wavy arguments, saying the test had to be done 'out of the box' and that turning off nodatacow introduced 'issues of data integrity'. https://www.phoronix.com/scan.php?page=article&item=eoy16-ws-server&num=1 would i assume phoronix tests are authoritative without investigation? - not likely. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On Fri, 16 Mar 2018 14:37:49 +0100 nicholas cunliffe <ndcunliffe@gmail.com> wrote:
to discuss sloppiness of tests (or at least an unwillingness to clarify test assumptions)
(its a while ago, but i believe this is the outline). in 2016 he did comparisons of different distros, testing read write speed using a database. After pointing out in the comments that in fact opensuse, btrfs guides and (i think) the authors of the database recommend using nodatacow for such applications the comment was 'set upon' by michaels fanboys using lame, handy wavy arguments, saying the test had to be done 'out of the box' and that turning off nodatacow introduced 'issues of data integrity'.
https://www.phoronix.com/scan.php?page=article&item=eoy16-ws-server&num=1
would i assume phoronix tests are authoritative without investigation? - not likely.
Hmm, methinks you overstate your case, which helps noone. There's only one fanboi commented on what I presume to be your comment, and he asked what seems to me to be a very sensible and pertinent question: "shouldn't opensuse set up the database data directories correctly by default?" Especially given openSUSE default choice of btrfs and extensive use of subvolumes, that seems like a reasonable thing to do. Perhaps rather than criticising the test, submitting a change request for the openSUSE package would be a more productive use of your time. BTW, it was Leap 42.2 or 42.3, not clear which, that was apparently tested, not TW. Also BTW, which test assumption wasn't stated and/or did he refuse to clarify? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Hmm, methinks you overstate your case, which helps noone. not really, ill try to make my point clearer:
'a reported analysis WITHOUT a basis of test which states or clarifies critical controlling parameters is essentially drivel' There's only one fanboi commented on what I presume to be your comment, not really: - i would say that statements like "so it's not as easy as "just tune each distro to the max away from defaults to be fair"" are not a rational reply to a fundamentally important technical question. All of the responses are 'defensive' rather than 'debative'. - statement constructs (which try to exaggerate the original question) like 'just tune each distro to the max' are often used to be 'dismissive' when not wanting to argue on technical merit. - 'Michael measures out of the box' i.e. using his first name and 'speculation of "michael's" intentions' rather than a neutral discussion of the document and facts fuels speculation of potential bias.
and he asked what seems to me to be a very sensible and pertinent question: "shouldn't opensuse set up the database data directories correctly by default?" canonical directories for intensive IO (like databases and VMs) are set up as nodatacow by default. I have no idea where the phoronix test suit installs to, or in which directory it performs IO tests.
Especially given openSUSE default choice of btrfs and extensive use of subvolumes, that seems like a reasonable thing to do. Perhaps rather than criticising the test, submitting a change request for the openSUSE package would be a more productive use of your time. specifying a change request to modify/create sub-volume/directories where a random user 'might' choose to install an application would be unlikely to succeed
BTW, it was Leap 42.2 or 42.3, not clear which, that was apparently tested, not TW. irrelevant to the argument - i am discussing test methodology, reporting, and attitude/response to critical evaluation.
Also BTW, which test assumption wasn't stated and/or did he refuse to clarify? if the directory in which the database was installed had the nodatacow setting (it can be fundamental to performance).
-- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
NB PLEASE DON'T COPY ME ON YOUR REPLY. I DO READ THE LIST. On Fri, 16 Mar 2018 20:33:14 +0100 nicholas cunliffe <ndcunliffe@gmail.com> wrote:
Hmm, methinks you overstate your case, which helps noone. not really, ill try to make my point clearer:
'a reported analysis WITHOUT a basis of test which states or clarifies critical controlling parameters is essentially drivel'
Not sure how that's relevant?
There's only one fanboi commented on what I presume to be your comment, not really: - i would say that statements like "so it's not as easy as "just tune each distro to the max away from defaults to be fair"" are not a rational reply to a fundamentally important technical question. All of the responses are 'defensive' rather than 'debative'. - statement constructs (which try to exaggerate the original question) like 'just tune each distro to the max' are often used to be 'dismissive' when not wanting to argue on technical merit. - 'Michael measures out of the box' i.e. using his first name and 'speculation of "michael's" intentions' rather than a neutral discussion of the document and facts fuels speculation of potential bias.
It would be a lot easier to read if you had properly quoted what I wrote. I don't see how any of what you wrote above addresses my comment that there was exactly one responder to your comment, not more than one.
and he asked what seems to me to be a very sensible and pertinent question: "shouldn't opensuse set up the database data directories correctly by default?" canonical directories for intensive IO (like databases and VMs) are set up as nodatacow by default. I have no idea where the phoronix test suit installs to, or in which directory it performs IO tests.
Especially given openSUSE default choice of btrfs and extensive use of subvolumes, that seems like a reasonable thing to do. Perhaps rather than criticising the test, submitting a change request for the openSUSE package would be a more productive use of your time. specifying a change request to modify/create sub-volume/directories where a random user 'might' choose to install an application would be unlikely to succeed
BTW, it was Leap 42.2 or 42.3, not clear which, that was apparently tested, not TW. irrelevant to the argument - i am discussing test methodology, reporting, and attitude/response to critical evaluation.
I didn't say it was relevant. I'm just correcting an error in your subject line. That's why it is a BTW.
Also BTW, which test assumption wasn't stated and/or did he refuse to clarify? if the directory in which the database was installed had the nodatacow setting (it can be fundamental to performance).
But YaST installs the database on openSUSE. As arjan_intel said, its up to the distro to make sure that it gets the settings right. More fundamentally: - the test article doesn't mention databases anywhere - it all happened over a year ago Why are we discussing this now, and what exactly are we discussing? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
- we are discussing the validity of phoronix tests. this has been the principle response to the original post. - one way to discuss if tests are 'for entertainment purposes only' (as said earlier) or 'authoritative' is the analysis of a particular case. - The case i have highlighted performs test using intensive IO (database or not) for which it is widely known to use nodatacow on btrfs. - The phoronix test does not specify the IO disk location or alternatively specify if nodatacow was used --> hence the test is essentially useless. i will make a simple analogy - if a magazine tested a 4wd car offroad and did not specify if 4wd was actually engaged would the results of the test be of any use? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
participants (2)
-
Dave Howorth -
nicholas cunliffe