Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 7 Oct 2014 16:43:30 -0600
From:      Stephen Mcconnell <stephen.mcconnell@avagotech.com>
To:        dgilbert@interlog.com, Borja Marcos <borjam@sarenet.es>,  Alan Somers <asomers@freebsd.org>
Cc:        FreeBSD-scsi <freebsd-scsi@freebsd.org>
Subject:   RE: mpr vs mps performance
Message-ID:  <b128bf97c211606b2a750e94f3b6df35@mail.gmail.com>
In-Reply-To: <54343A0F.6080905@interlog.com>
References:  <A8A5F62D-ED7B-420F-B5B5-88842A8F8827@sarenet.es> <CAOtMX2jVR-8wkHcTyudOVwMNmxOvDkWPNsTBuyu%2BWZQ6WtDvmg@mail.gmail.com> <c7d0c371bb7db87e433040f198102297@mail.gmail.com> <CAOtMX2h5O7EUf4H-oewPy1aAZ%2BYrwdRZLqr-=S2EbE6wjerAeg@mail.gmail.com> <BD011874-CA6D-42AB-846C-CFAEA734A867@sarenet.es> <54343A0F.6080905@interlog.com>

next in thread | previous in thread | raw e-mail | index | archive | help
> -----Original Message-----
> From: Douglas Gilbert [mailto:dgilbert@interlog.com]
> Sent: Tuesday, October 07, 2014 1:08 PM
> To: Borja Marcos; Alan Somers
> Cc: Stephen Mcconnell; FreeBSD-scsi
> Subject: Re: mpr vs mps performance
>
> On 14-10-07 12:19 PM, Borja Marcos wrote:
> >
> > On Oct 7, 2014, at 6:11 PM, Alan Somers wrote:
> >
> >>> Borja, can you upgrade your card to the latest FW.  Your version is
> >>> really old (1.00.03.00).  We should be up to 7.00.00.00 now.  There
> >>> could be a lot of fixes that would affect performance.
> >>
> >> To update the firmware, you must download LSI's
> >> Installer_P6_for_FreeBSD from the link below.  Then the tricky part:
> >> you must hexedit the executable and replace all "mps3_" strings with
> >> "mpr", suitably padded with NULLs.  It sucks, I know; complain to
> >> LSI/Avago.  Then you must download the
> >> 9300_8e_Package_P6_IT_Firmware_BIOS_for_MSDOS_Windows (or
> whichever
> >> is appropriate for your hardware) to get the actual firmware files.
> >> Then you can use the hacked FreeBSD installer binary to download the
> >> firmware from the Windows firmware package.
> >>
> >> http://www.lsi.com/support/pages/download-results.aspx?component=Stor
> >>
> age+Component&productfamily=Host+Bus+Adapters&productcode=P00766&as
> se
> >> ttype=0&productname=LSI+SAS+9300-8e
> >
> > Thanks, I'll try. Probably the SAS2 cards will go into production but
I'm happy
> to help with some tests.
> >
> > The performance problem was not sustained, but writing performance
stalled
> for several seconds now and then. It did not happen with the SAS2 HBA
(which is
> a Dell branded as H200 flashed to IT firmware).
>
> I'm not seeing a big difference between FreeBSD and Linux using a LSI
SAS-3 HBA
> with a 12 Gbps SAS SSD.
>
> With the same 9300-4i HBA (P6 firmware (where is P7 that Stephen
mentioned
> ??)) and the same disk (SEAGATE ST200FM0073 0005), using Linux as the
client:
>
> Via iSCSI against FreeNAS 9.3-M4 (nightly build, 3 days old):
> # ddpt --odx if=/dev/sg5 bs=512 of=/dev/sg5 seek=4m count=4m
> 4194304+0 records in
> 4194304+0 records out
> time to transfer data: 11.275797 secs at 190.45 MB/sec
>
> On Linux lk 3.17.0 against the same disk, directly connected:
> # ddpt if=/dev/sg2 bs=512 of=/dev/sg2 seek=4m count=4m
> 4194304+0 records in
> 4194304+0 records out
> time to transfer data: 10.807345 secs at 198.71 MB/sec
>
>
> The SSD is probably not optimized for an "up-copy" like that. With raw
READs it
> yields better than 700 MB/sec, using both ports it yields better than
900 MB/sec.
>
> "--odx" is a Microsoft term for a token based remote copy. It was
proposed to
> T10 as "XCOPYv2: Extended Copy Plus & Lite" and the result is found in
SBC-3.
>
> Doug Gilbert

Yesterday I mentioned that most of the performance gains on the Gen3 card
come from MSI-x support.  In fact, MSI-x Firmware support does add to
performance, but I forgot about the actual hardware I/O acceleration in
the Gen3 cards (Fast Path I/O), which increases performance significantly.
Just wanted to add that.

Also, I did a little testing and I got about the same numbers as Doug with
the 3008 and a 12Gb SAS SSD (a little better than 700 MB/sec).  With the
2008, I saw raw reads of around 425 MB/sec.  I don't do a lot of benchmark
testing, but that's what I saw in my little test.

Steve
>
>
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b128bf97c211606b2a750e94f3b6df35>