I don't have an issue with them requiring a controller lift to enable the file persona. My assumption (and perhaps it's a poor one), is that the reason it took so long for HP to add file persona is that they wanted to do it right, and in order to do it right, it took more resources than they had available on the hardware. The other thing is that prior to the HP acquisition, the original 3par crew was very adamant about not wanting to run additional features like file persona, AO, etc on the node. they wanted to minimize the load on the node CPU/Memory as much as humanly possible. So it took awhile for the culture to change enough where the engineers would be ok with adding that feature.
At the end of the day, features like file persona, dedupe, etc are never especially innovative in and of themselves. What I love about 3par is that hp has added these features without even requiring an offline upgrade, which imo is impressive. That being said, compare how HP has done file with EMC and Netapp (the other big players in the converged storage arena), and I see a number of advantages. (warning, I'm about to get into marketspeak. My opinions are my own. I do not work for HP or EMC or Netapp, but some of HP's marketing has really rung true for me)
In EMC VNX, despite the fact that it's "converged" it's really only converged in the management. You still need the file heads (celerra) to support file. In EMC VNXe, the file head runs as a VM, and required dedicated disks. Yep that's right folks, you can't store file and block on the same set of disks. There's a bunch of other stuff about VNXe that makes me nuts, that being one. EMC does the same thing with flash cache. You have to dedicate disks to flash cache, and they cannot also be used as a tier for primary data.
In Netapp, the whole thing runs file natively, and they added a block emulator. This means netapp is great for file, but block gets the short end of the stick.
In 3PAR, the file persona runs on top of block (the way it should be. Why on god's green earth would you put a block emulator which writes to file which get's translated back to block to go out to the disks netapp? WHY?). We can share the disks between block and file with no issues. All the goodies we know and love like inline dedupe, zero detect, thin provisioning, AO, DO, replication apply to file the same way they to file that they do to block, and we don't need any additional hardware or VM to run file (I'm looking at you EMC).
On it's surface, it may not seem especially innovative, but HP seems to have implemented this feature the right way, and the details seem to me anyway to be pretty innovative. And while it's disappointing, I can understand the need to do a controller lift to get it. As far as online, I will admit that my head might explode of HP can controller lift a 7200/7400 online. I mean seriously, that would be impressive in a huge way. Even more so when EMC can't firmware upgrade xtremeIO online!
|