On Thu, Sep 27, 2012 at 9:41 PM, jason@xxxxxxxx <jason@xxxxxxxx> wrote: > One thing that comes to mind for me is your 1gb test may be mostly hitting > the ram cache on the ssd. That is generally where ssd makers, consumer and > to a lesser extent enterprise, get the peak I/O ops numbers from. I understand your thinking. But, in tests using Facebook FlashCache as well as our internally developed caching software (I work for a big storage company), that is not what I see. What I see is that once the cache warms ups, regardless of the size (i.e. 1GB, 10GB, 100GB), that access times are sub-millisecond. > Can you do the same tests, 10gb and 1gb, using only the ssd as your block > device to get a baseline for it without bcache in the picture? Good suggestion! I have done this and the data shows that my SSD has a baseline of 800K iops when using a 4k blocksize. > I would also do some mixed r/w tests just to sort of get a real world > profile for your ssd and system. I'll take a look at doing this over the weekend. > You also don't mention what type and how many spindles you have in the > mechanical array. I'm using a RAID array which has lots of horsepower and the RAID 5 configuration has 6 drives. Each drive is a 15K rpm type. So the array has very good performance. One data point that is when I run Facebook flashcache, once the cache is fully warmed, I see about 0.60 ms response time. Any suggestions? -brad w. -- To unsubscribe from this list: send the line "unsubscribe linux-bcache" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
