Anyone got any experience of deploying OE on a SAN which dynamically allocates disks? SSD, SAS, and SAS near line storage all available, and data is dynamically stored in relation to how it is accessed. The data more often used is moved to SSD drives where it can be accessed quicker. The software is an EMC VNX on VMware.
This means that allocating BI and AI files across different volumes is not possible unless separate LUNS are assigned for them and while I am no expert in this area I am told that solution is messy. I am also told there is a mixed RAID solution in this and RAID 5 may be in use but I have not found out if this is the case or where it would kick in.
The application is experiencing performance problems when the user count increases and several DB tuning, software improvements and some infrastructure (upping DB server RAM, changing SAN disk provisioning from thin to thick) issues are being implemented/considered.
I am wondering if the EMC VNX solution might be causing problems and if dedicated disks with full mirroring would be a better solution.
In the modern HW environment from a performance perspective SAN storage is pretty much always inferior to internal disks. If performance is your goal then it is awfully hard to argue against internal SSD. Which is also probably cheaper.
Think about it... SAN storage is external and shared. Both of those facts contribute major obstacles. They only way that SAN can win is to be compared to crippled internal disks. 20 years ago that is in fact how SAN got a leg up. You could not physically put enough internal disks in a server to match the storage capacity or the IO Ops of a SAN. The landscape today has changed. Capacity stopped being a challenge ages ago and SSDs cured the IO Ops problem. Now the latency from server to SAN is a significant chunk of the IO response -- there is an order of magnitude difference in an IO to an internal SSD vs an IO Op to a SAN. If your external storage is also shared then you're in even more trouble.
I have had some experience with EMC VNX and it is not an exception. In the last round of testing that I did on one of those when we put all of the data on SSD it was actually *slower*. The speculation was that there was a bottleneck on the SSD controller but we didn't keep digging into it.
> On Jun 22, 2016, at 6:09 AM, conroypt wrote:
>
> changing SAN disk provisioning from thin to thick
Good to get rid of thin provisioning.
The way thin provisioning works is like this:
You create a 100 GB file system on the SAN.
The SAN lies to you and says ok, you have 100 GB free.
You write something to the filesystem and the SAN goes: oh, crap, he wants to write something but no space has actually been allocated and I have no place to put this data. I better go get some space.
Then the write completes.
Lather, rinse, repeat.
Then again, sometimes the SAN lies to 10 people and each thinks they have 100 GB on 500 GB of actual storage. On day 1, that may work. Eventually, the 500 GB is used up and everybody thinks there is plenty of free space. Guess what happens then?
Thanks Tom. I hadn't come across these SANS with dynamic storage before and would never have recommended them for Progress anyway. I guess from the couple of responses here they are not widely used. The latency you mention from server to SAN and the delays from thin provisioning outlined by Gus (thanks very much Gus) could well be contributing to the poor performance. I will post any relevant info.