<   Previous Post  SATA Performance
So just what do I...  Next Post:   >

Comentarios (5)

1 localhost ha hecho un comentario el Retroenlace

I didn't know about Datacore- the only other in-band, non Hitachi solution I've come across is Falconstor.<div>&nbsp;</div> As for the relative strengths and weaknesses, I would regard the code quality and performance of the solution more important than whether or not the hardware is provided by the same vendor. All Intel servers are created equal, more or less ;)

2 localhost ha hecho un comentario el Enlace permanente

Yes, I missed Falconstor, should have included. <div>&nbsp;</div> My point about the platform was more about the base operating system used and eliminating as much of it as possible. Most operating systems have big fat layers between you and the actual hardware registers and DMA buffers. With the approach used by SVC we basically pull the data direct from the HBA and post it back onto the HBA - eliminating device driver and other software layers. Without this approach, you cannot guarantee the latency overheads.

3 localhost ha hecho un comentario el Enlace permanente

Actually there is only a partition license on the DS4000 line, not on the DS8000 line or the NSeries line.

4 localhost ha hecho un comentario el Retroenlace

Hi Barry, thanks for details.<div>&nbsp;</div> I am sitting here pondering the value of additional nonvolatile storage cache in a storage network device, especially in a world where many arrays offer the same, and read caching is a standard part of most host OSes.<div>&nbsp;</div> Could it provide some additional performance enhancement in some situations, e.g. really dumb storage attached?<div>&nbsp;</div> Perhaps, but at a cost.<div>&nbsp;</div> One cost is the additional hardware and logic required for nonvolatility. Another, more subtle cost is the inability to use array-based replication to make consistent, point-in-time copies.<div>&nbsp;</div> Another thread to this discussion is around economic fit. As an example, a key theme to storage virtualization is "use your old storage". Given that most IT assets are rolled every 2-3 years for economic reasons, are there certain use cases where the customer would just be better off putting multiple tiers of new storage inside the array?<div>&nbsp;</div> I think one useful discussion is "when does external storage virtualization make sense" vs. "when does inside-the-array pooling" make sense.<div>&nbsp;</div> Many of the benefits you elicit can be found with either host-based storage virtualization (e.g. volume managers), or virtualization within a larger array.<div>&nbsp;</div> As far as "simplified management", I think that can be debated, as on one hand any decent SRM product will help you show and change what you've got -- no appliance required. And on the other hand, SVC specifically does not appear to offer an end-to-end management capability of host, HBA, switch, etc. It does, however, do a nice job of hiding the underlying storage.<div>&nbsp;</div> Thanks

5 localhost ha hecho un comentario el Enlace permanente

Some interesting points Chuck, and a few ideas for future topics in there.<div>&nbsp;</div> I plan to discuss the caching advantages etc in my part-work underway. As for copy services, as long as the appliance can provide the same or better level of copy services then you haven't lost any ability. Again, something I'm drafting now for the next part.<div>&nbsp;</div> Simplified management could be a contentious issue, as you say SVC does not provide end to end management, but it was one of our design points, SVC is not a SAN manager, its a virtualization appliance, and it virtualizes the storage, so needs to do a good job of hiding the underlying storage.