The new nodes are very different, it's not just a cache DIMM swap as has been pointed out. The new controllers have more powerful processors (7200 went from 4 cores to 6 cores, 7400 went from 6-core to 8-core... 7450 stayed the same, 8 cores, but higher clock speed I believe), and larger SSD drives for boot/cache storage space.
Incidentally, someone mentioned that they wish there was an option to get more cache in an existing 3Par 7xx0, such as the 2x cache sizes that are available in the new 7xx0c "NAS-enabled" models.
Well, although it doesn't appear on price lists and hasn't been messaged out to limit adoption, for the first time ever, 3Par actually has come up with a "data in place" controller node-swap upgrade for existing 7200s, 7400s, and 7450 arrays to replace the existing nodes with the newer, double-the-cache "c-suffix" model nodes.
It's not free, basically what you end up doing is buying a new pair or two new pairs of nodes, through your normal reseller partner and the distribution channel just as if you were buying an upgrade (presumably with roughly the same discounting that applied when you purchased your 3Par to start with), and HP does have a supported way to swap out the controllers/nodes with the newer double-cache nodes.
The array has to go down, unlike if a single node has to be replaced, and this is one reason 3Par does not like to do "data in place node upgrades" - 3Par prefers to consider their arrays as never having to incur planned downtime. But this is a special case. In reality, the "upgrade" of both nodes happens in a similar way to a single node replacement, except that the 3Par OS cannot tolerate two nodes in the same node pair having unlike processors or cache sizes. For a typical node replacement, the CPU and Cache/RAM will be identical, and the OS is set to accept a new/replacement node of like class and add it into the cluster of nodes. Since both the processors and cache are larger (more cores and more cache), no on-line upgrade is possible, the OS would reject the new node if you tried it, IIRC the new node will simply fail self-test as an incorrect replacement part and never be added to the cluster.
All this to say, for a fairly small upgrade price (the nodes are NOT a very large cost in these arrays, and any FC or 10GbE cards can be moved to the new nodes, so no additional cost there), you CAN get the larger cache sizes, and the enhanced processors (more cores, and I think it's a clock speed tick higher as well) in your existing 3Par 7xx0 "non-NAS nodes" array. You'll need to request it through your normal reseller, whoever you get your 3Par and/or upgrades from, and they may get back a "this doesn't exist" on their first query to their distributor. Trust me, it DOES EXIST, it's just not well know and has only been on a very few presentations. It's the only time 3Par has allowed this type of in-place upgrade, so it will probably take pushing the entire way to move people past their complacency and "we've never done it that way" mentality... but it can be done.
FWIW, the other posters hit most of the high points on the original question about whether AFC is worthwhile. You have the right approach with the try-it "what if" reporting from within 3.2.1, and that should be all you need to get a sense of whether AO is doing the job for you or if AFC might help. Actually, though, they complement each other, since AO does a great job but only after 4 or 8 or 24 hours of trend analysis, and AFC moves active data into AFC read cache after its first use, right away, and keeps it there for a while waiting for it to be used again.
Both "speed up" methodologies have their place in the scheme - AFC can move new data into Flash for fast reads, and AO cannot do that. So, if you have new data landing that needs to be in Flash as well as existing data that AO has already moved to Flash, that's where handing some Flash to AFC makes sense. And almost everybody has SOME new data coming in from apps, the web, or whatever. That "2%" improvement might spike briefly to a 20% improvement from time to time as new data comes in and needs to be on a higher tier. You're looking at long-trend averages, consider how new data comes in to your database app or whatever you're hoping to accelerate.
Take Care and Good Luck.
|