1) You don't need any battery powered controllers with HA. Your UPS acts as a battery. The same about controller cache. It's a day before yesterday technology... Your whole machine acts as a storage controllers so system RAM is used for caching. Skip expensive RAID models and throw money into more SAS or SATA ports and more system RAM.
2) Go for cheap high-capacity SATA hard disks in RAID10 instead of expensive SAS ones in RAID5/6.
3) CPU horsepower does not matter much. More cores/sockets are preferred to high clock number however.
4) More RAM you're going to put into your HA nodes - more distributed write-back cache we'll allocated. Good number is 80% of your free system RAM.
With proper MPIO performance scales nearly linearly. So from 4 GbE links I'd expect ~400MB/sec for read or write and more if reads and writes are combined. But you'd need V5.8 (not released yet) for full speed. V5.6 and V5.7 we're going to release soon (V5.7 = V5.6 + bugfixes actually) have a lot of spin locks put to keep data safe (we don't confirm write before it would not touch both hard disk platters on both HA nodes) so for versions up to V5.7 HA performance is 50%-80% of non-HA performance. With V5.8 and up you'll see multiplication of number of links and nodes.
petr wrote:Hi all,
we are planing to use starwind for our Hyper-V cluster. Now we have about 6 - 10 VPS servers each with (16 - 32 GB RAM) and we are running Hyper-V virtual servers on them. I would like to move to iSCSI solution. Could you help me with configuration (choose right parts)?
My first idea - each starwind HA node will contain:
+ Supermicro 3U 16 HDD chassis E1 model wid SAS expander
+ Adaptec 6805 with Flash backup module (600)
+ 16x WDC RE4 1TB or 2 TB hdds
+ Intel Xeon (don't know which to choose)
+ 16 or 32 GB RAM ECC
Every node will have 2-4 1Gbps connection each to another switch and to each starwind HA node. How fast will be this configuration with MPIO enabled?
Do you have any recommendations?
Thank you
