Hi All,
I read the article here http://wikibon.org/wiki/v/The_IBM_XIV_Storage_Array_Performance_and_Availability_Envelope
It mention the maximum available IOPS is calculated using the following...
The overall maximum application I/O rate to disk is 60 x 180 x ½ (mirroring) x 90% (other disk overhead) = 4,860. If the cache hit rate is 85%, this represents an overall I/O rate of 32,000 IOPS
I can see the math of 4860 IO, but how does the cache hit rate math part work?
Update: I think the closest I can get to is taking 4860 *( 1-85%) = 32400 .
Can someone explain to me what this means? Does that mean the system only needs to access the disk 15% of the time, thus make the cache that much more useful?
Thanks
Hi Tamagoliu
Your update is nearly correct - it should be 4860 / (1-85%) = 32,400. Yes, that means that 85% of the time the array will find the data in cache, and 15% of the time the array will have to fetch the data from disk storage.
David Floyer Wikibon CTO
Thanks David. For the answer. Does this apply to all type of Storage Array calculations? Regardless the cache size?
I know some storage system, such as Compellent does not use a large amount of cache memory as a buffer, in that case , do we look at pure physical disk IO for it's performance?
Jay
Hi Jay
The relationship between cache size and hit rate is a complex one. It varies by type of workload (sequential for example will tend to bypass cache) and the specific vendor implementation. Very generally, doubling the cache size will halve the miss rate% (1 - hit rate%).
Lower end controllers tend to be Intel servers and the cache size is constrained by the base server maximum RAM size and the capabilities of the cache backup system (usually battery based, but increasingly flash is used). The high-end storage arrays have much larger cache sizes, and much higher prices. Compellent has the Model 30 controller with 3.5GB cache (battery backup) and the Model 40 with 4GB (backed up with flash). A 2-controller array has 8GB of flash total, compared with up to 32GB at the high end of the EMC CLARiiON and Hitachi AMS arrays. Compellent has a fully virtualized architecture and higher functionality.
In general for larger latency sensitive database workloads with poor locality of reference, you are likely to need additional cache. For more ordered workloads with a requirement/desire for additional array functionality, Compellent will be excellent.
Let us know what you decide!
David
Hi David,
Thank you for the explanation on the cache responds time information. We already bought one Model 40 with 4 GB cache. We are running mix workload, database, web services, etc. We are seeing some increase in latency as the IO (mostly random) increases. I really think some additional cache would be helpful.