Il giorno 27/ago/2010, alle ore 21.46, Mark ha scritto: > Right now we're using NFS for our ESX Servers. The only iSCSI LUN's I have > are mounted inside a couple Windows VM's. I'd have to migrate all our VM's > to iSCSI, which I'm willing to do if it would help and not cause other issues.
It won't change. VMWare issues synch writes to both NFS and iSCSI so you'll gain nothing and you'll lose the flexibility of NFS. But since the "problem" lies in an "abuse of synch writes" (asynch writes from the guest OS becomes synch writes to VMWare datastore) I suggest you to do one of 2 things: 1. move your data to NFS/CIFS. Mount the very same NFS server (of course different dataset) inside the guest OS and put your data there. PRO: optimized network trafic (as ops will be "to the byte" and not "to the block") and reduced (albeit not zero, since NFS still synch file close etc, I don't know about CIFS) "synch abuse", you can access you data from outside the VM easily. CON: may not be possible depending on the kind of application, you're bound to the network file protocol semantics which can be not enough. 2. move your data to an iSCSI disk. Export a zvol to a guest OS, format it and move your data there (mind volblocksize). PRO: no "abuse of synch writes": no asynch writes will turn into synch writes as the guest OS will have complete control over the device. CON: less optimized network traffic (should be more or less what you see with vmdks), possibly performance/stability issues as the support for iSCSI is not yet mature in certain OSes (ie: https://bugzilla.redhat.com/show_bug.cgi?id=583218) "abuse of synch writes" has a PRO, tho: it makes your data safer in case of crash or power loss. Back in the day before caches all writes were inherently synchroous and a programmer knew that after a write returned the data was safe. Nowadays it's not the case anymore and the coder should wisely use synch writes or other ways to do things to ensure safety of data. Battery backed RAM based SLOGs gives you the best of both worlds, but they cost money :) -- Simone Caldana _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
