Opera running on SSD hard drives

richardd

Does anybody have OperaII or Opera3 running on SSD hard drives in the server? We have a customer wanting to improve speed / disk access.

colin-avey

Hi Richard

My in house server has four SSD drives in a RAID 5. I once did some performance testing for one of our customers to prove issues with their infrastructure. STRAN, CTRAN and NTRAN are arround 80%plus capacity on this site.

The test was a bit of bespoke which imports XML orders from their web site as SOP documents via the developer import hooks. I have just copy and paste the relevant part from the document below. If want to know any more get back to me.

Colin
------

I have run some test imports of BluBolt Web orders and taken various timing from the log files. I think the results tell an interesting story. Ray for your benefit I have included a copy of the invoice for the bits for what is essentially a home built file server. It is running vSphere5. The vm for the fileserver is 6 cores & 24GB RAM. It is running SBS 2011 with just about every option loaded. Background IOPs are peaking at about 250/second. I used Perfmon to monitor at various stages, both the client and the server. The whole server has been configured for power saving not performance, although the RAID controller is a serious bit of kit. I have a 1GB network single link only active

I did tests both from my office desktop and a VM on the same host as the server. Vm on server 4 cores & 4GB RAM.

As I suspected network traffic went up massively once the 2nd user connected. Difficult to judge, but looked like 5 or 6 times as much.

Batch size : 252 Orders

Run from my desk top this took 44:48 – or between 10 and 11 seconds per order.

Run from VM on 38:50 – or nearly 9 seconds per order.
This produced peak Disk IOPs of 6941/sec >82MB/sec on the server and network IOPs on the client of 6774/sec

I had a log at the log to see if there were any standout out slow areas. I have picked up one area in particular which can be improved. The import was originally designed to expect unique IDs for delivery addresses from BluBolt as it gets for the main account. The code a safety function to cope with people who didn’t want to register where if no ID came from BluBolt then the system would search for the next available number. This routine is not very optimised as it was never expected to get heavy usage, however it would appear you are not get IDs from BluBolt and this routine is being used all the time. In fact there is a note in the code and a mod indicating I was asked to ignore the ID on all orders at a later date.

I decide to do another test to try and ascertain the impact of this. I remove the above mentioned line of code and manually inserted a unique ID in to orders.

Run on VM 27:14 – or 6.5 seconds per order
This produced peak Disk IOPs of 1022/sec >78MB/sec on the server and network IOPs on the client of 2559/sec

I believe these result are still ultimately effect by IO limitation of my small setup as the VM never exceeded 10% processor utilisation. I would expect any less than 25% in this setup would indicate IO as the limiting factor.