Dave Hitz hitz@netapp.com writes:
It remains to be seen whether or not we're crazy, but perhaps I can convince you that we're not *completely* crazy.
I understand your skepticism, because running the entire backup data flow through Java would probably not result in acceptable performance.
Our idea is to use Java to execute the Legato daemons, and to direct the data movement, but to provide new primitives in WAFL that allow Java to say things like: "stream the data from this inode to this network connection." I'm not sure what all WAFL primitives Jim ended up adding to boost performance, but hopefully you get the idea. We're not talking about the normal Java I/O library here.
So Java orchestrates the backup, but the data flow itself goes directly from WAFL to the network, using the same basic code path as NFS, CIFS or HTTP.
Dave,
This seems like marketing-deal-driven engineering.
A backup system that requires every client to run a specialized daemon (which must be ported every time you want to use a new type of client) seems really broken to me. We're quite happy here using dump, restore, and (nowadays) NDMP.
I'm worried that these additions will not only cause us problems on our filers, but will encourage Legato to kill off BudTool.
Most of all, I don't really understand the effort being applied to Java/Legato when other simple, useful, and important protocols like SNTP, SNMP, and rsh are unsupported or neglected.
- Dan