December 10, 2012 at 4:56 pm
My first post so hello to everyone for the first time 🙂
I work for a company who make a large ERP product and my role is to spec hardware for the onsite Managed Service i.e. we manage from hardware up but in the customers data centre. There are around 10 non DB Server VM's and 2 SQL Server VM's for Production and a Dev that hosts multiples of Production database.
Currently we use the following systems
VMWare ESXi 5.1 + vCenter
HP DL380 x 2 | 64GB | Quad Core Xeons x 2
HP iSCSI P2000 | 10k for VM OS Drives and 15k for SQL Server 2008
GB Cisco Switches x 2
I am now working on an IBM solution and for a similar cost I can get the same spec with a few less disks in the SAN and 1 x 365 GB Fusion I/O card for one of the hosts. I have a 765GB one on loan at the moment and wow its fast but that's just in ATTO and IOMeter tests..
Some of our customers have 200GB DB's plus BLOB tables stored on slower disks.
Do you think it would be better to fill the VM Hosts with loads more memory and allocate this to the virtualised DB server or to stick with 64GB and use a Fusion I/O card to host the database?
1 x 8GB DIMM = £75
1 x 16GB DIMM = £160
1 x 365GB Fusion I/O card = £3508
If my memory allocation is bigger than my database what are my bottlenecks? CPU and Disk Writes?
If you have any ideas on hardware HP or IBM then please let me have them 🙂 Go wild...
December 12, 2012 at 4:56 pm
Couple of things you may want to look at. First is how the licensing on your VM software works out with the change in ram allocation. Second is redundancy with the fusion I/O drive, what are you going to do if that single component fails?
"If my memory allocation is bigger than my database what are my bottlenecks? CPU and Disk Writes?"
What does the existing load look like through your entire I/O profile? I've seen enough poorly designed vendor applications that this question is likely going to be very specific to your environment. "It depends", especially throwing this on VM, reservations, over provisioning and balloon drivers now go into the mix.
It may be worth it for you to make a call to someone like Brent Ozar's team, or just spend some time looking at some of the vast amounts of free information he's published online.
December 13, 2012 at 4:17 am
Hi thanks for the response
The design of the hardware is for future customers of ours so I am looking into how to mitigate performance risks. We use VMWare and have 2 hosts plus a SAN but we only host our software on it and only 10 - 15 VM's run on this kit. We already over spec the hardware to take into account 5 years of growth and unknown factors.
The software we make and sell is very customisable and each customers performance profiles will be different so as before I want to make get the best bang performance within our budget for new hardware.
I am worried about the single point of failure of the card but we can get around this with either 15 minute log backups or SQL Server 2012 Always on or DB Mirroring.
Basically I want to spec the best hardware for around £35k.
December 17, 2012 at 5:22 am
I would not have an important live server relying on a single FusionIO card.
RAID1 is the safe route for production.
Got several of these cards in various configurations - example 2x640 in RAID0 create a pretty fast 1.2TB drive:-D, but if you cannot afford service disruption it's best to have them as a single 640GB in software RAID1. Yes it sucks that you 'waste' a card and the throughput is a bit lower, but you are doing the right thing:cool:
Also remember to allow for free RAM just to handle the FusionIO cards, they need RAM space to 'breath' (used to be ~2GB RAM for every 80GB of SSD, YMMV)
So there you have it, you cannot skimp on RAM after all:-)
Cheers,
JohnA
MCM: SQL2008
Viewing 4 posts - 1 through 3 (of 3 total)
You must be logged in to reply to this topic. Login to reply