Erez Zilber | 2 Dec 10:54 2007

[PATCH]: add missing ';' in iscsi_exec_abort_task

Mike,

There's a missing ';' in linux-2.6-iscsi git tree in the "linus" branch.
This bug was fixed in the "iscsi" branch in commit
0a35c9b479ded645aa1a51547af9cd9aa8bf22db, but it still exists in
upstream code.

Erez

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi@...
To unsubscribe from this group, send email to open-iscsi-unsubscribe@...
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

TECB | 2 Dec 14:07 2007
Picon

Re: Performance problem running Open-suse 10.2


Hi. After some more investigating.. seems that the XFS FS is
bonkered... so mounting it crashes the server... I think we can rule
out the iscsid for now.

On Nov 30, 1:55 pm, TECB <dragm...@...> wrote:
> oki, today ive reinstalled the whole os from scratch to suse 10.3.
> Everything went fine til i mounted the iscsi box ... the whole server
> just froze.. The 3 leds on the keyboard started blinking ;)
> iscsid -v = 2.0-865 Opensuse 10.3
>
> On Nov 29, 9:31 pm, TECB <dragm...@...> wrote:
>
> > After a reboot and reconnect to the iscsi box i got :
> > scsi_transport_iscsi: no version for "struct_module" found: kernel
> > tainted.
> > but i was able to start iscsid, mount and use the volume.. I even got
> > over 100MB/s performance..yay!! But copying a 4gig file.. and 50% the
> > box crashed.. coulndt see anything in the logs
>
> > On Nov 29, 6:32 pm, Mike Christie <micha...@...> wrote:
>
> > > TECB wrote:
> > > > Hi. Hmm problem is that i dont get the /dev/sdb1 up any more after
> > > > running the init script..
> > > > gettting a :
>
> > > > storage1:/ # /etc/init.d/open-iscsi start
> > > > Setting up iSCSI targets: iscsiadm: No records found!
>
(Continue reading)

TECB | 2 Dec 17:16 2007
Picon

Re: Performance problem running Open-suse 10.2


Just an update on my findings for u that are interested.
It seems that going from Suse 10.2 to Suse 10.3 wasnt a smart idea..
The inbuilt XFSprogs (2.8.21) and the manually compiled 2.9.4.1 can
both mount and xfs_repair my 3TB 3ware volume with XFS, tho when
trying to even xfs_repair or mount the 10,5TB iscsi xfs volume just
bombs the server. Im baffled to say it the least.. never seen this on
any distribution or kernel before.. Im manually compiling 2.6.23-9 now
to check if the xfs modules there would be any better... Does anyone
know if the iscsi modules in 2.6.23-9 are compatible with the 865-15
Userspace tools thats default in suse 10.3 ?
PS! remember that this FS worked fine in 10.2 and earlier distribution
versions..

best regards tecb

On Dec 2, 2:07 pm, TECB <dragm...@...> wrote:
> Hi. After some more investigating.. seems that the XFS FS is
> bonkered... so mounting it crashes the server... I think we can rule
> out the iscsid for now.
>
> On Nov 30, 1:55 pm, TECB <dragm...@...> wrote:
>
> > oki, today ive reinstalled the whole os from scratch to suse 10.3.
> > Everything went fine til i mounted the iscsi box ... the whole server
> > just froze.. The 3 leds on the keyboard started blinking ;)
> > iscsid -v = 2.0-865 Opensuse 10.3
>
> > On Nov 29, 9:31 pm, TECB <dragm...@...> wrote:
>
(Continue reading)

TECB | 2 Dec 19:36 2007
Picon

Re: Performance problem running Open-suse 10.2


Last update ;)
2.6.23-9 did the trick.. I was able to mount, umount and repair the
FS.
Open-iscsi worked fine and the performance is roughly 90-110MB/s on
read with blockdev -setra 16384, and rougly the same on write..
Im not getting the stellar 110-123MB/s i had on the 765 builds i had a
year ago.. but its better than 40-60mb/s.
So the conclusion is; Suse 10.3, Large XFS volume on iscsi = bombs on
the default kernel.

best regards te

On Dec 2, 5:16 pm, TECB <dragm...@...> wrote:
> Just an update on my findings for u that are interested.
> It seems that going from Suse 10.2 to Suse 10.3 wasnt a smart idea..
> The inbuilt XFSprogs (2.8.21) and the manually compiled 2.9.4.1 can
> both mount and xfs_repair my 3TB 3ware volume with XFS, tho when
> trying to even xfs_repair or mount the 10,5TB iscsi xfs volume just
> bombs the server. Im baffled to say it the least.. never seen this on
> any distribution or kernel before.. Im manually compiling 2.6.23-9 now
> to check if the xfs modules there would be any better... Does anyone
> know if the iscsi modules in 2.6.23-9 are compatible with the 865-15
> Userspace tools thats default in suse 10.3 ?
> PS! remember that this FS worked fine in 10.2 and earlier distribution
> versions..
>
> best regards tecb
>
> On Dec 2, 2:07 pm, TECB <dragm...@...> wrote:
(Continue reading)

Roger Bystrøm | 2 Dec 20:04 2007
Picon

Re: open-iscsi and jetstor 416is


Thanks! I'll look into those.

This is my first go at disk arrays and so I'm terribly inexperienced.
What we need is raid6 or preferably raid60 and at lest 6-7 TB storage
capacity and a couple of hot spares that we can put in a 3U.
Personally I'd really like to use open-iscsi to connect to the box so
we don't have to spend lots of $$ on HBAs.

The jetstor box looks great, but absolutely no info about it and
linux/open-iscsi on the net :/

Roger

On Nov 30, 2007 10:45 AM, Tim <storagegeek007@...> wrote:
>
> Hello Roger,
>
> There are couple more products storbank-xl and Integrated Storage
> Concentrator which are similar to these products, check them out at
> stonefly.com and dnfstorage.com, I had worked earlier on these
> products and they seem to be much reliable and user friendly till
> date. Let me know your exact requirement and setup of environment so
> that I can suggest you some other vendor products apart from these in
> the market as per your requirement.
>
> :)
> Tims
>
>
(Continue reading)

Pasi Kärkkäinen | 3 Dec 07:52 2007
Picon
Picon

Re: open-iscsi and jetstor 416is


On Sun, Dec 02, 2007 at 08:04:10PM +0100, Roger Bystrøm wrote:
> 
> Thanks! I'll look into those.
> 
> This is my first go at disk arrays and so I'm terribly inexperienced.
> What we need is raid6 or preferably raid60 and at lest 6-7 TB storage
> capacity and a couple of hot spares that we can put in a 3U.
> Personally I'd really like to use open-iscsi to connect to the box so
> we don't have to spend lots of $$ on HBAs.
> 
> The jetstor box looks great, but absolutely no info about it and
> linux/open-iscsi on the net :/
> 

http://www.equallogic.com/products/view.aspx?id=1791

works well with Linux / open-iscsi.

-- Pasi

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi@...
To unsubscribe from this group, send email to open-iscsi-unsubscribe <at> googlegroups.com
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

benjamin9999 | 3 Dec 19:54 2007

Re: Protocol Error on packet recieved from initiator.


On Nov 30, 9:37 am, maguar887 <mforou...@...> wrote:
> I do not see any errors in demsg when this happens.

small net, big greetings to you sir

wireshark has a decoder for iscsi protocol which might be of use

run iscsid with more verbose debug detail ( i usually run in
foreground, -f -d5 or so ) and you may catch some more detail - but
probably target is disconnecting the array because it's not happy with
the "error on packet" so you may only see open-iscsi reconnecting for
"no error"

and FWIW, i've had a lot of trouble with e1000's.  i don't know if
there were bad production runs or what, but i have seen two different
symptoms where replacement card (of same type) fixed everything.  one
caused CRC errors in the iscsi stream - even with data digests..., and
more frequently i get TX-hang, device resets where the the connection
will timeout while the card is unresponsive during this cycle.  as
best as i can tell, it's faulty boards/ports (i'm using e1000 MT dual
boards, 82546GB.)

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi@...
To unsubscribe from this group, send email to open-iscsi-unsubscribe@...
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

(Continue reading)

Simone Morellato | 4 Dec 04:55 2007

RE: iscsi Performance

We are running IOMeter dynamo traffic, 512B sequential reads on a
PowerEdge 2950 w/ Intel Xeon 5110 (4 cores), 1G ram. With
 
Test Setup 1) 10G card: Intel PRO/10GbE SR Server
Test Setup 2) 4x1G cards Intel PRO1000 Server
 
Intro:
The performance in linux is not influenced by the number of outstanding IOs and it increases with the number of targets/dynamos
 
The performance in windows is instead influenced by the number of outstanding IOs and from the number of target/dynamos
 
Windows vs Linux:
 
On the "4x1G setup" Windows and Linux performance with 1 outstanding IO is comparable, but with 16 outstanding IOs windows performance is 3 times higher.
 
On the "10G setup" Windows performance is almost double than Linux when only 1 outstanding IO is used, and since Linux performance does not change with the number of outstanding IOs, Windows performance becomes almost 4 times better when 16 outstanding IOs are used.
 
10G vs 4x1G:
 
Now let's compare Windows and linux performance in the 2 setups.
 
Windows performance numbers are very similar in both "4x1G" and "10G" setups.
 
Linux performance numbers on the "10G setup" are almost half than the "4x1G setup".
 
 
Please take a look at the attached spreadsheet with the data collected.
Thanks,
Simone
 
 
-----Original Message-----
From: open-iscsi-/JYPxA39Uh5TLH3MbocFFw@public.gmane.org [mailto:open-iscsi-/JYPxA39Uh5TLH3MbocFFw@public.gmane.org] On Behalf Of Ming Zhang
Sent: Friday, November 16, 2007 5:27 PM
To: open-iscsi-/JYPxA39Uh5TLH3MbocFFw@public.gmane.org
Subject: Re: iscsi Performance
 
 
On Fri, 2007-11-16 at 13:09 -0800, Simone Morellato wrote:
> Hi Mike,
>
> Why do I get 4 times more performance if I use 4 1G nics instead of a 10G nic? I am doing IOMeter 512 bytes reads.
> Can I do something to get the same performance of the 4 1G nics with a single 10G nic?
 
 
i bet nobody can answer this question if you did not give detail HW/SW
configuration, iometer workload setting, ...
 
 
>
> Thanks,
> Simone
>
>
> >
--
Ming Zhang
 
 
<at> #$%^ purging memory... (*!%
--------------------------------------------
 
 
 
 

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi-/JYPxA39Uh5TLH3MbocFFw@public.gmane.org
To unsubscribe from this group, send email to open-iscsi-unsubscribe-/JYPxA39Uh5TLH3MbocFFw@public.gmane.org
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

Attachment (PerformanceData.xls): application/vnd.ms-excel, 23 KiB
Pasi Kärkkäinen | 4 Dec 08:21 2007
Picon
Picon

Re: iscsi Performance


On Mon, Dec 03, 2007 at 07:55:56PM -0800, Simone Morellato wrote:
> We are running IOMeter dynamo traffic, 512B sequential reads on a
> PowerEdge 2950 w/ Intel Xeon 5110 (4 cores), 1G ram. With
> 
> Test Setup 1) 10G card: Intel PRO/10GbE SR Server
> Test Setup 2) 4x1G cards Intel PRO1000 Server
> 
> Intro:
> The performance in linux is not influenced by the number of outstanding IOs and it increases with the
number of targets/dynamos
> 
> The performance in windows is instead influenced by the number of outstanding IOs and from the number of target/dynamos
> 
> Windows vs Linux:
> 
> On the "4x1G setup" Windows and Linux performance with 1 outstanding IO is comparable, but with 16
outstanding IOs windows performance is 3 times higher.
> 
> On the "10G setup" Windows performance is almost double than Linux when only 1 outstanding IO is used, and
since Linux performance does not change with the number of outstanding IOs, Windows performance becomes
almost 4 times better when 16 outstanding IOs are used.
> 
> 
> 10G vs 4x1G:
> 
> Now let's compare Windows and linux performance in the 2 setups.
> 
> Windows performance numbers are very similar in both "4x1G" and "10G" setups.
> 
> Linux performance numbers on the "10G setup" are almost half than the "4x1G setup".
> 
> 
> Please take a look at the attached spreadsheet with the data collected.
> 

If I remember correctly IOMeter had some bug which prevented number of
outstanding IO's parameters working under Linux ? 

And in that case your bad results would be (partly) caused by IOmeter being
buggy under Linux? 

-- Pasi

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi@...
To unsubscribe from this group, send email to open-iscsi-unsubscribe@...
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

Pasi Kärkkäinen | 4 Dec 08:55 2007
Picon
Picon

Re: iscsi Performance


On Tue, Dec 04, 2007 at 09:21:52AM +0200, Pasi Kärkkäinen wrote:
> 
> On Mon, Dec 03, 2007 at 07:55:56PM -0800, Simone Morellato wrote:
> > We are running IOMeter dynamo traffic, 512B sequential reads on a
> > PowerEdge 2950 w/ Intel Xeon 5110 (4 cores), 1G ram. With
> > 
> > Test Setup 1) 10G card: Intel PRO/10GbE SR Server
> > Test Setup 2) 4x1G cards Intel PRO1000 Server
> > 
> > Intro:
> > The performance in linux is not influenced by the number of outstanding IOs and it increases with the
number of targets/dynamos
> > 
> > The performance in windows is instead influenced by the number of outstanding IOs and from the number of target/dynamos
> > 
> > Windows vs Linux:
> > 
> > On the "4x1G setup" Windows and Linux performance with 1 outstanding IO is comparable, but with 16
outstanding IOs windows performance is 3 times higher.
> > 
> > On the "10G setup" Windows performance is almost double than Linux when only 1 outstanding IO is used, and
since Linux performance does not change with the number of outstanding IOs, Windows performance becomes
almost 4 times better when 16 outstanding IOs are used.
> > 
> > 
> > 10G vs 4x1G:
> > 
> > Now let's compare Windows and linux performance in the 2 setups.
> > 
> > Windows performance numbers are very similar in both "4x1G" and "10G" setups.
> > 
> > Linux performance numbers on the "10G setup" are almost half than the "4x1G setup".
> > 
> > 
> > Please take a look at the attached spreadsheet with the data collected.
> > 
> 
> If I remember correctly IOMeter had some bug which prevented number of
> outstanding IO's parameters working under Linux ? 
> 
> And in that case your bad results would be (partly) caused by IOmeter being
> buggy under Linux? 
> 

Also, can you tell use more about your Linux configuration? 

Did you configure/increase (tcp) memory buffer/window sizes? What other
options did you tweak? 

Especially 10G speeds require some tweaks to get good performance. 

-- Pasi

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "open-iscsi" group.
To post to this group, send email to open-iscsi@...
To unsubscribe from this group, send email to open-iscsi-unsubscribe <at> googlegroups.com
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---


Gmane