Re: Dell Perc 5/i Performance issues

From: Scott Long <scottl_at_samsco.org>
Date: Wed, 30 Jun 2010 11:17:35 -0600
On Jun 19, 2010, at 10:31 PM, Garrett Cooper wrote:
> On Sat, Jun 19, 2010 at 7:56 PM, Scott Long <scottl_at_samsco.org> wrote:
>> On Jun 19, 2010, at 5:32 AM, Garrett Cooper wrote:
>>> On Sat, Jun 19, 2010 at 4:24 AM, Garrett Cooper <yanefbsd_at_gmail.com> wrote:
>>>> On Sat, Jun 19, 2010 at 4:21 AM, oizs <oizs_at_freemail.hu> wrote:
>>>>> Since I tested it on different kind of os's, and with at least 5 testing
>>>>> applications, I don't think that would be the case.
>>>>> 
>>>>> On 2010.06.19. 13:17, Garrett Cooper wrote:
>>>>>> 
>>>>>> On Sat, Jun 19, 2010 at 2:58 AM, oizs<oizs_at_freemail.hu>  wrote:
>>>>>> 
>>>>>>> 
>>>>>>> I tried almost everything raid 0 1 5 10 with all kind of stripes
>>>>>>> 32/64/128
>>>>>>> and settings direct io/cached/read-ahead/wt/wb/disk-cache but nothing
>>>>>>> seems
>>>>>>> to work.
>>>>>>> I changed the card to another dell perc 5 which had an older firmware.
>>>>>>> Tried
>>>>>>> 4 kind of motherboards even tried changing the os to linux and windows
>>>>>>> xp/7.
>>>>>>> In windows I got some funny results 1.3MB/s with write-back and 150MB/s
>>>>>>> reads with 5 disks in raid0.
>>>>>>> I just wanted to have a hw raid with no problems since the motherboard
>>>>>>> 88sx7042 and bsd did not like eachother.
>>>>>>> 
>>>>>>> On 2010.06.19. 11:07, Svein Skogen (Listmail Account) wrote:
>>>>>>> 
>>>>>>>> 
>>>>>>>> On 18.06.2010 01:50, oizs wrote:
>>>>>>>> 
>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> Hi,
>>>>>>>>> 
>>>>>>>>> I've bought a Dell Perc 5/i because I couldn't make the onboard marvell
>>>>>>>>> 88sx7042 work with 8.0/8.1 or current, but as lucky as I am, the best I
>>>>>>>>> can do with 4x1.5tb samsung in raid5 is 60MB/s writes and 90MB/s reads,
>>>>>>>>> with bbu/write-back/adaptive-read-ahead.
>>>>>>>>> 
>>>>>>>>> I was expecting at least twice of that, and I'm not sure what can I do
>>>>>>>>> to get that speed. (I've read man 7 tuning with no success)
>>>>>>>>> 
>>>>>>>>> As far as I know this controller should be as fast as on other systems.
>>>>>>>>> (Freebsd.org mx1 has one of these cards.)
>>>>>>>>> 
>>>>>>>>> I'm hoping somebody on the list reads this and helps because I can't
>>>>>>>>> afford to buy another card.
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>> 
>>>>>>>> I've lost track of what actual boards Dell has OEMized to make the
>>>>>>>> various PERCs, but if I remember somewhat correctly, the PERC5 is
>>>>>>>> basically an LSI Megaraid SAS 8308elp with different labels and
>>>>>>>> firmware?
>>>>>>>> 
>>>>>>>> If so, I've got that exact controller (minus the dell labels and
>>>>>>>> firmware) in my primary storage box here, and yes, you SHOULD be able to
>>>>>>>> get more performance out of it. What's your strip sizes and logical disk
>>>>>>>> layout?
>>>>>>>> 
>>>>>>>> (I've got the same board running on 8x 1T5 Seagates in RAID5+0, and that
>>>>>>>> setup easily pulls 5 times the values you're seeing, and by all logic
>>>>>>>> you should see about half of what I'm seeing)
>>>>>>>> 
>>>>>> 
>>>>>> Dumb question: are you sure that the problem that you're seeing isn't
>>>>>> in fact inhibited by the application that you're getting `performance'
>>>>>> results with?
>>>> 
>>>> If your applications aren't well suited for your hardware's
>>>> capabilities, then of course performance will be bad.
>>> 
>>> Furthermore, if the performance applications and your use scenarios
>>> are centered around reading, as opposed to writing, there is an option
>>> within mficontrol and the mfi(4) interface where you can actually
>>> enable read-ahead, instead of writeback (you unfortunately can't
>>> enable both scenarios). I realize that this is an artificial
>>> improvement in a way, but you should judge whether or not your
>>> application will be doing more reading than writing in whatever
>>> capacity it's doing...
>>> 
>>> HTH,
>> 
>> No, that doesn't help.  I wrote the driver, and I have no flipping clue what you're talking about.
> 
>    Nevermind. It was a misunderstanding of what the subcommands...
>    - mfiutil cache .. enable
>    - mfiutil cache .. reads enable
>    - mfiutil cache .. writes enable
>    ... do.
> -Garrett

I'm very late in doing this, but I wanted to publicly apologize to Garrett both for being rude and flippant to him, and for doing so in public.

Scott
Received on Wed Jun 30 2010 - 15:17:40 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:05 UTC