new raid1 code, speed improvements!! :D please benchmark it

[rspadim]

hi guys, i made some changes to md raid1 software, could gentoo community test
it? for me it work very nice =)
the raid1 new code is based in kernel 2.6.37
here is the new and old code:
www.spadim.com.br/raid1

just read_balance changed,4 modes:
near_head(today)
round_robin(counter per mirror)
stripe (like raid0, with shift for make stripe less or more intensive)
time based (depending on head positioning time and read size it calculate the
fasted mirror, some new improvement must be done but it works, improvements=
get io queue of each mirror and make a time estimation of queue time, with it
get more close best disk selection)

it´s open source please help me testing it, neil at raid kernel need some
benchmarks to put it inside kernel source

for mixed speed mirrors time based is good
for ssd only round_robin is good (per mirror count, put 230 for 230mb/s, 150
for mb/s or multiples to make a good round robin)
for disks and/or ssd mixed or not, stripe is good (like raid0)
for disks only near head is good
thanks

[Telemin]

Hi, can I please make couple of suggestions to you, I mean no offence in any of the below but I feel that since your post has gone largely ignored for the last two days that I might be able to help with suggestions even though I cannot test it myself (I have no raid array:( ).

First of all, since we don't actually know who you are, and the code comes only from a mirror on your site how can we be sure it is safe to use on our own raid arrays? Although nobody in their right mind would ever test anything on a production server we still value the stability of our machines, and few of us have the luxury of dedicated test boxes.

To this end, and since few of those in a position to test have the time and expertise to trawl through the code to see what you have changed, could you please give us a description of how and why you have changed what you have changed so testers will feel secure that the code is safe to run.



Secondly, the current way you have made the code available again means it is not easy to test without knowing how to apply the changed code (as far as I can tell it is not even a patch) so could you please consider:

a) making it available as a patch version rather than single files, and

b) writing some brief instructions on how to correctly patch and configure


Thirdly, precisely what benchmarks do you require? I can think of a reasonable number of more and less scientific ways to perform the benchmarking, and if you want comparable results from everyone you will have to specify precisely what benchmarks you want everyone to run.

I don't want you at all to feel that I am preaching to you, as that is not at all the intention. This post is intended to help you get more people in to test, by clarifying questions they will have and making it easier for them to give you what you need.

I personally feel that people will be put off from helping you if they believe that it will be a lot of hard work, or will cause them problems with their system, and that many of the unanswered questions that will put people off are taken care of in the answers to the questions I have given you above.

Best of luck, and I hope you get the results you need.

-Telemin-
_________________
The Geek formerly known as -Freestyling-
When you feel your problem has been solved please add [Solved] to the topic title.
Please adopt an unanswered post