Hitachi AMS 2100, 2300 and 2500 Performance Metrics
Does anyone know how to calculate max number of IOPS over a single port for an AMS 2100, 2300 and 2500 or there would be supported. I have read about IOPS here for USP and gave around 2500 IOPS per port. I am assuming that I may get similar numbers for AMS2000 series arrays. It might sound like a newbie question but I really don’t have any idea about it.
Re: Hitachi AMS 2100, 2300 and 2500 Performance Metrics
If I concerned about the front ports only, there are poor indicator for performance. AS I know the variation of blocks sizes greatly affect the IOPS number, which can be sustained by a frontend port. Sometime you might have noticed that frontend ports are bottleneck. In that case I urge you for the average tag count. If you get something around 512, then it will create problem for you.
Re: Hitachi AMS 2100, 2300 and 2500 Performance Metrics
In my opinion the very poor applications so far I have seen is Microsoft Exchange. Even if you get billion IOPs and Exchange would find some way screwing up FEP performance. I really experienced it in my case. I was using the Solaris system with Oracle using VxVM. At that time I was getting 1500 IOPs per USB port. At that time system was going ballistic. While backup phase, often seeing 4000 IOPs on the ports but it wasn’t much a problem. However during the day time, it was completely different.
Re: Hitachi AMS 2100, 2300 and 2500 Performance Metrics
Actually my intention is very much different here. I want to know how many servers I can map to a single port. I need information something about fan out ratio. I know that the above information you had given is very helpful for me.
To say, when I was on AMS 2300, I did notice the IOPS go as high as 2400 on a single port. On average it was always 1600 IOPS.
Re: Hitachi AMS 2100, 2300 and 2500 Performance Metrics
When the system is under mixed workloads, the IOPS between 1000 to 2000 being the breaking point. One thing sure that values totally depends on the block size. MS Exchange/SQL uses 8K blocks but the data warehouse may use massive blocks. You need to divide the ports based on workload profiles. I hope that in your setup, you have these functionalities.