From: wk <wit...@op...> - 2008-11-26 14:29:51
|
Vladislav Bolkhovitin wrote: > wk wrote: >> Vladislav Bolkhovitin wrote: >>> Hi, >>> >>> wk wrote: >>>> Hi Vladislav, >>>> >>>> Our QA team has tested several scenarios with changing parameters of >>>> iSCSI virtual disks - like changing size of virtual disk through lvm >>>> tools (on the iSCSI target side). >>>> They noticed that with IET and Windows as iSCSI initiator it is >>>> possible to resize logical volume - and some programs for Windows >>>> doesn't lose transmissions started before. But with SCST iSCSI some >>>> of these programs *always* loses its transmissions. With this patch >>>> I've sent - the QA team says that now SCST iSCSI works like IET >>>> (from the Windows user point of view) - so it doesn't break the >>>> connection. >>>> >>>> For doing such operations as resize we need of course to detach >>>> virtual disk from iSCSI target, resize it and attach it to LUN again. >>>> >>>> Here is the scenario of testing: >>>> >>>> 1. configuring iSCSI target (SCST or IET) with some VDISK >>>> 2. on the Windows side - configuring disk with iSCSI inititator >>>> 3. windows: - starting copying very big file >>>> 4. detach of VDISK on iSCSI target - from IET of SCST >>>> 5. reconfiguration of VDISK on iSCSI target side (i.e. resizing) >>>> 6. attach just reconfigured VDISK on iSCSI target again >>>> >>>> The steps 4-6 are managed by script - so they are relatively fast. >>>> >>>> Probably there is somewhere the problem - during reattachment of >>>> disk - and as effect Windows applications looses its connections? >>>> >>>> >>>> My patch in fact doesn't remove the original code of generation SCSI >>>> id - it only allows to set your own id, but if you don't set it >>>> manually - it will use the original method. >>> It's hard to believe that the current SCSI ID generation code can >>> cause such consequences, because the generated ID is always the same >>> for the same device name. I believe, the cause is in the >>> corresponding INQUIRY handling code. I'm currently fixing the similar >>> issue, so let's return to it after I finish. >> >> It seems like I'll have to explain the whole situation. >> >> We needed the possibility of setting our own SCSI ID in iSCSI. We are >> using some kind of failover solution where there is required identical >> SCSI ID on failover nodes. For this purpose patch I've sent earlier >> has been done. >> During the comparison testing of two iSCSI implementations: IET and >> SCST it seemed like IET was able to keep their connections up with >> initiator, but SCST worked properly only with our patch. It might be >> very particular situation that SCST with our patch worked fine without >> losing connections, because few days later we cannot repeat those >> tests with the same results: SCST were losing their connections >> independent of we were using original SCST or patched one. So, the >> conclusion that changing SCSI ID will help with keeping up established >> connections was wrong. The problem lies somewhere else. > > This is what I meant. But questions about ability to set manual SCSI ID > as well as USN keep rising again and again. This is why I suggested to > return to your patch after the code it touches will be proved to be > correct. Then it would be good if your patch also will be extended to > allow to set USN. > >> Thank you for the answers, and sorry for the fuss. >> >> p.s. for testing we were using "sequential tests" from "Bart's Stuff >> Test" programm on the Windows system. > > I'm not aware about it. Can to send me the link, please? I'm looking for > good small test for Windows. Hi Vladislav, Below is the direct link for "Bart's Stuff Test" we were using for tests: http://www.nu2.nu/download.php?sFile=bst514.zip best regards, Witold Kowolik |