> > Hi There,
> >
> > I'ld like to use an iSCSI-Target for my
> TimeMachine
> > Backup.
> > So I take Comstar with Default settings and the
> > globalSAN iSCSI Target Software for MAC.
I have been using Comstar iSCSI target on various snv and oSol respins for
quite some time. It seems to me one has to be very careful matching the target
and the initiator. I've had very good experience using various versions of the
Microsoft iSCSI initiator to different targets (some commercial SANs, Solaris,
both Comstar and the original daemon, a lot of NetBSD boxes used only as iSCSI
targets), the results unfortunately vary a lot with the initiator.
> >
> > Problem: Transfering Data ist to slow. (1GBit
> > Ethernet). Took about 10 hours for 50GB what SMB
> does
I've posted this elsewhere... On a Gigabit network with Jumbo frames I am
presently trying to install OpenSolaris on a VirtualBox machine, hosted on
W2K8R2 system, using the built-in VirtualBox initiator, the target being a
SX:CE system, snv_125 BFUed to 128. Up to 20% of the installation goes
reasonably fast, then the speed drops a lot, I start seeing disk errors in the
virtual machine (I ran format->analyze->verify on a 15GB iSCSI disk, it took me
some three days with about 150 bad blocks reported; on the target there were no
indications of an error on the physical disk at all).
After the format I decided to try the installation once more. In two days I am
on 80% done with about 4480 error messages in the log file of the type:
...
Error for command 'Write sector' - error level fatal
Sense key: ID not found
Vendor 'Gen-ATA ' error code: 0x05
...
with very low bandwidth usage on the target (and very low iostat on the
underlying ZFS pool -
...
[cheeky] ~ # zpool iostat bpool 1
capacity operations bandwidth
pool alloc free read write read write
---------- ----- ----- ----- ----- ----- -----
bpool 24.5G 32.5G 45 45 2.59M 320K
bpool 24.5G 32.5G 68 0 4.07M 0
bpool 24.5G 32.5G 74 0 4.56M 0
bpool 24.5G 32.5G 70 0 4.31M 0
bpool 24.5G 32.5G 66 0 4.13M 0
bpool 24.5G 32.5G 70 0 4.38M 0
bpool 24.5G 32.5G 54 501 3.44M 2.67M
bpool 24.5G 32.5G 50 122 2.67M 1.90M
bpool 24.5G 32.5G 60 10 3.36M 82.9K
bpool 24.5G 32.5G 81 2 4.78M 24.0K
bpool 24.5G 32.5G 60 11 3.26M 95.9K
bpool 24.5G 32.5G 61 12 2.92M 104K
bpool 24.5G 32.5G 69 10 3.21M 87.9K
bpool 24.5G 32.5G 60 12 3.81M 104K
^C
....
(no other acivity on this pool ATM, only the single zvol Comstar target).
> > (on the same Server) in something about an hour.
> >
> > Could someone help me tuning this up?
This was just an example what one can hit using iSCSI in untested combinations.
Caveat emptor, as usual.
> >
> > Thx
> > Henri
> >
> > PS: Added the same post at Network-Forum a few
> days
> > before (without an answer). Hope thats OK
>
> I can't help, because I'm running the old iSCSI
> daemon rather than the
> new COMSTAR based target. But it does work, and
> perhaps faster than
> you've experienced, although it's still rather slow.
> (I think I also ported one
> f the BSD iSCSI target daemons to Solaris 9; it also
> appears to work with the
> GlobalSAN initiator last I tried, but I didn't push
> it much.)
As I mentioned above, I have been using NetBSD targets for quite some time.
Again, it depends a lot on the underlying infrastructure. While I was on a
100MB network with an old iprb card, I had no problems whatsoever; I had even
left the NetBSD box on an old 4.99.51 version. Since I switched to a Gigabit
network, I've hit a number of problems (especially with Jumbo frames, which I
am gradually beginning to understand).
>
> One of these days, I'll update my SB2K to SXCE 129,
> and then maybe I'll
> be able to try this sort of thing, although to really
> push it will have to wait
> until I get more disks.
With the exception of the machine I mentioned above, all my OpenSolaris
machines follw the latest respins and are on 130 ATM. There is one setup as a
target in a W2K8R2 Hyper-V fault-tollerant cluster and SCVMM which has never
had any problems doing it's job (the speed is also as expected).
> (a GigE would be nice too,
> but I don't think I have any
> slots left)
Chavdar Ivanov
--
This message posted from opensolaris.org
