StarWind iSCSI SAN Version 8.0 Beta-2

[fbifido]

URGENT !!!!

http://www.starwindsoftware.com/forums/ ... tml#p19921


Can i request a few features :
if any feature does not make any sence, mybad.

1) Send Beta Log ( betalogs@xxxxx.xxx)
2) Send log file (support@xxxxxx.xxx)
3) Repair/check/scandisk lsfs device
4) mount/unmount lsfs device
5) Anyother commands that can be use on lsfs devices

6) Have an option to create a temp lsfs disk LUN to do speed test and creaste a client side software to do the test from within a VM or on any hypervisor. client side test software for xp, vista, win2k3/2k8/2k12, win7/8 and does I/O test.

7) same as "6" but with lsfs RAM LUN. use same client side software. ALso have the option to see what the perf. is with the starwind feature on and with it off.

check for UPDATE, as a menu item, in managemt software. Must do live-update/patching

9) We need a little more info on how the starwind iSCSI SAN service consume memories, like what settings affect how must memory is to be used & how these settings affect perf., Can we calculate the max needed... etc ....

10) Can you make the software knows the speed of disk that it will use to create the lsfs files. and make the Statistical info part of the core-back-bone, as in:
it knows the speed of things, HD, RAM, NICS, CPU
it knows the speed of transit, Data-In, Data-out, Data-dedup, Data-to-HD, Data-from-HD, time-of-each-Data-path, bottle-neck-of-Data
It knows about time, time-to-complete each action, time everything.

[fbifido]

iSCSI SAN not working

Log files attached ...

[fbifido]

I just removed the problem LUN's from my Target and esxi is now seeing my Target and iSCSI SAN is now stable again.

How can I add these problem LUN's to esxi so that I can move my VM's out to another LUN?

[Vitalii (staff)]

You may try opening the needed LSFS device as read-only, so you will be able to read all the data from it.

To do this, edit the .swdsk file of that device. It has an XML format.
find the following line:

Code: Select all

<mode>1</mode>

and replace it with

Code: Select all

<mode>2</mode>

then try to mount this device, it should be mounted as read-only.

[barrysmoke]

I have a complex setup, so bear with me. I'll be testing different things to find the problem:
vmware 5.1 update 1
vsphere cluster, with 2 esxi hosts, each host has dual port 10 gig cards, with crossover cable, and a vswitch1 with vmkernel on 172.16 network for my iscsi traffic
host1 has an lsi card in passthrough mode to a windows 2012 standard vm(not r2...I'll test that next), on which I installed v8 beta2 link from this thread(starwind management console version 8.0.6049)
each lsi card has 20x 2.5" disks attached through a sas expander, although for my first test, I'm just using 8 disks as a comparative benchmark to my stormagic setup.

step1 was going to be to test creating an iscsi san datastore, and pass that to vmware, format vmfs-5, and run an iometer benchmark off of the datastore.
I was able to create the target, add device, created a virtual disk on e:(lsi raid drive), enabled maximum l1 cache(32 gigs on the vm), and l2 cache stored on c:(which in actuality is a virtual hard drive for this vm, although this is the only vm on a raid1 mirrored ssd)

each vmware hosts sees the iscsi target on initial creation. the first host creates the vmfs-5 datastore from the target, and I browse datastore, and upload an iso file, so I have some files to look at, and know each hosts can upload data to the iscsi target.
here's where things go wrong. on the second host, I add a disk to one of my other vm's hosted on local storage. this vm will be a 100gb disk, to be stored on the newly created starwind iscsi datastore. at the end of the disk creation, which runs for a really long time...longer than I have ever had take(thick-eager zero)....I can't see any files on the datastore. it's completely empty. you can't upload any files to it anymore. it is still mounted.

here is the log file:http://bsmokeman.no-ip.biz/fileshare/12 ... 005011.log

[barrysmoke]

worked by passing the lsi raid disk straight through(diskbridge) I didn't enable any caching...so it might be caching, I'll run some benchmarks, then reconfigure with virtual disk, no caching to narrow it down.

[barrysmoke]

While I had the diskbridge in place, I ran some benchmarks, which reflect what anton was saying about performance hit on passing direct hardware disks as iscsi target.
the problem is, they are severely worse than the same config with stormagic. I moved on to re-create the virtual disk san, without any caching. That was successful, no data loss, as experienced previously with caching on.
That allowed me to run more iometer benchmarks, and while there is an improvement over the diskbridge, these are still much worse than my stormagic benchmarks.
I'm off to find my bottleneck, and I'll even be downgrading to 2008 server, and v6 just to make sure.

lots of testing in my future. I am concerned about the data loss with the cache enabled, has anyone else seen that? could just be me trying to use cache inside a virtual machine. that should work with a memory reservation though.

[barrysmoke]

also to go along with the cache issue, I tried creating a ramdisk of 90 gig, and passing that to a vmware host, to format vmfs.
it created successfully, but a vmfs format doesn't work, unless you make the formatted size smaller than "maximum available".
then, once created, it's useless, you can't upload any files to it, or browse, etc...
again I suspect using the vm ram as a ramdisk is what's causing this behaviour.

would upgrading to windows server 2012 r2 make any difference?

what about running v8 beta2 on windows 2008r2?

[barrysmoke]

I think I found the bottleneck. It's either a previously known passthrough issue with the lsi 9271-8i, and vmware 5.1, or it is an issue with firmware on my intel oem chassis. those are the only 2 things that are different, and I'm getting much better results testing on my second chassis.

I'll update here when I find out for sure.

[anton (staff)]

People tend to do this again and again so I guess we'll disable mapping raw disks as a targets at some point...

While I had the diskbridge in place, I ran some benchmarks, which reflect what anton was saying about performance hit on passing direct hardware disks as iscsi target.
the problem is, they are severely worse than the same config with stormagic. I moved on to re-create the virtual disk san, without any caching. That was successful, no data loss, as experienced previously with caching on.
That allowed me to run more iometer benchmarks, and while there is an improvement over the diskbridge, these are still much worse than my stormagic benchmarks.
I'm off to find my bottleneck, and I'll even be downgrading to 2008 server, and v6 just to make sure.

lots of testing in my future. I am concerned about the data loss with the cache enabled, has anyone else seen that? could just be me trying to use cache inside a virtual machine. that should work with a memory reservation though.

[Bohdan (staff)]

The build has been updated.
http://starwindsoftware.com/tmplink/StarWindBeta.exe
Could you please also check the update process from v6 to v8? All devices types should work properly, but Snapshot&CDP, Mirror and DDdisk devices have restricted management functionality. There may be issues with LSFS in synchronous replica configuration. Single LSFS devices should work properly, as well as LSFS snapshots.
Update from previous beta builds may not work correctly.
Also we are interested in performance results for Synchronous replica devices based on thick images.

[turkina]

The build has been updated.
http://starwindsoftware.com/tmplink/StarWindBeta.exe
Could you please also check the update process from v6 to v8? All devices types should work properly, but Snapshot&CDP, Mirror and DDdisk devices have restricted management functionality. There may be issues with LSFS in synchronous replica configuration. Single LSFS devices should work properly, as well as LSFS snapshots.
Update from previous beta builds may not work correctly.
Also we are interested in performance results for Synchronous replica devices based on thick images.

So we should not update our Beta2 build to this new build? We should perform a fresh install?

[Bohdan (staff)]

So we should not update our Beta2 build to this new build? We should perform a fresh install?

v8b2 update may be not smooth, so basically yes, either fresh install or update of v6 to this build.

[DUOK]

The build has been updated.
http://starwindsoftware.com/tmplink/StarWindBeta.exe

Hi Bohdan,

I do not understand. This is the same version as before (beta2 b6049). Is there a mistake?

[Bohdan (staff)]

Please try once again.