May 8, 2011 at 9:10 pm
Comments posted to this topic are about the item Tiered Storage
May 9, 2011 at 7:20 am
It's not just access patterns you need to be aware of when deciding where to put the data. You also need to keep in mind how quickly the data needs to be accessed when it's accessed. At my last job we had a client archive old scanned documents to a jukebox optical drive. The first time they tried to access any image on that system through our software it failed because it couldn't respond fast enough. Fortunately, if they went to load it again it loaded right away, probably because the disk was already loaded.
May 9, 2011 at 8:25 am
Speed definitely matters. If your software doesn't take a longer access time into account, it can be an issue.
May 9, 2011 at 12:43 pm
Tiering makes sense, if you can either handle a SAN doing it automatically (i.e. unpredictable performance), or you can split your workload up properly.
The article itself has a few issues, to my mind, particularly in the examples whose math is used to support the "faster and cheaper" argument:
1) SSD's in JBOD? They should be in RAID1, RAID10, or RAID5 to provide redundancy, just as traditional disks are. If you really insist on throwing away redundancy, at least use RAID0 and get some more speed out of it.
2) The example Tier 2 storage only held 192GB of the 250GB of "warm" data.
3) Less important, but still of note for those considering SSD's, is a completely lack of discussion of the type of IO; particularly on SSD's this is critical. A handful of SSD's in RAID5 may roughly match a score of SAS disks in RAID50 on random 64KB Writes, while completely trouncing that same SAS setup on random 64KB Reads.
I believe local SSD's are the way to go for certain IOPS and throughput sensitive workloads, but I still believe some disks/cables/trays/controllers/whatever will die from time to time. That said, one really needs to benchmark their SSD's; they're very sensitive to specific usage pattern, and tend to display throughput caps on your controllers, as well.
Has anyone experimented with using filtered indexes to put the hottest "part of an index" on a different filegroup?
May 10, 2011 at 5:59 pm
I like a couple of comments made about the tiered storage.
1) You should understand and know your data and data access patterns
2) Get to know the DMVs
I like the idea of tiered storage and it seems most appropriate for a database - if you understand the data. It can help to improve overall performance.
Jason...AKA CirqueDeSQLeil
_______________________________________________
I have given a name to my pain...MCM SQL Server, MVP
SQL RNNR
Posting Performance Based Questions - Gail Shaw[/url]
Learn Extended Events
Viewing 5 posts - 1 through 4 (of 4 total)
You must be logged in to reply to this topic. Login to reply