need config recomendations for small MD3220i and VMware
I'm a one man IT shop supporting about 100 desktops and an IT heavy business. Currently I have 3 esxi boxes on essencials plus licenses. I'm moving towards shared iscsi storage. The 3 servers share 12-15VM's including 100 exchange users, web hosting, SQL, and a handful of other light duty line of business servers. Looking for recomendations on how allocate storage. I ordered 8x300GB 10k disks. I'm not sure how to allocate this storage because so much of what I read online seems to be for enviroments larger than mine. I'm leaning towards one 1 RAID 5 disk group and a single LUN. Two disk groups seems dumb because i loose another disk to pariety (I can buy more though). I've been leading towards adding two 500GB NLSAS disks for sql and exchange log files just to have them on different spindles. For the network it seems recomended to have each port in its own subnet... no big deal. The servers (supermicro) each have 4 nics so I can put 2 to iSCSI use. Any opinions on direct connecting the servers to the switches? Its 100MB/sec able to do more than what it seems like?
ESXi's maximum LUN size is "2048GB minus 512 bytes", so your initial thoughts wouldn't work.
NLSAS is a glorified SATA drive; it has the same performance as a SATA drive, but it just happens to have a SAS PCB on the side (maybe a bit better testing and more reliable firmware, but the physical harddrive is the same as a SATA drive). It makes no sense to put your performance stuff on your slowest drives.
Last I checked VMware hasn't validated direct-connect iSCSI setups (only direct connect fiber channel or SAS), so for support reasons I would avoid that, but in theory if properly planned, it may work.
If you have the budget for an MD3220i with 8 x 300GB drives, 3 servers and a VMware Essentials plus license, I'd just do the setup right and get a set of decent iSCSI switches. If you want to stick with Dell, a pair of 5424's are more than enough (the 2800's aren't the best choice, the 5500 is overpriced right now, and the 6200, 7000 and 8000-series are way more than you need.
I'd also look into adding a dual port or quad port NIC to each server. The reason is that once you have shared storage, it would make sense to set up/enable vMotion. This requires a dedicated Gbit NIC (I usually use the same switches I use for iSCSI, but just create a separate VLAN on them for vMotion). So, with 4 NICs, you'd have 2 for iSCSI, 1 for vMotion and only 1 left for management and all your VMs to connect to your LAN... which lacks the redundancy that a setup like this deserves. If you add a dual port, I'd take 3 ports for management and VM LAN access (on vSwitch0 usually). If you add a quad port, I'd use 4 NICs for iSCSI, 1 NIC for vMotion and 3 NICs for management and VM LAN access.
I was leaning towards the NLSAS disks only to have SQL logs on different spindles, I don't think I'd do anything more than that with it. I'm now reading that exchange 2010 does not benifit from this like it used so, so I may not go that way.
I'll look into iSCSI switches. I've got 4 nic's now. I was thinking of two for iSCSI, 1 for vMotion, and one for managment and LAN access. I've not seen anyone else do this, but what would be the implications of sharing an iSCSI port with vMotion?
My big question remains though. Is there any down side to using a single disk group / single LUN for 3 hosts. I have the single controller version so I'm not selling myself short by not utilizing the other controller.
The problem with sharing one of the iSCSI NICs for vMotion is that your iSCSI bandwidth will be unbalanced while you're performing a vMotion migration. As the SAN uses Round Robin for the iSCSI multipathing, and vMotion can only be enabled on a single vmkernel at a time, it ends up causing a severe bandwidth reduction on both iSCSI connections when you're doing a vMotion.
I'd pick up a dual port card for each server (e.g. this Intel card) and not have to sacrifice/impact my setup so significantly.
Not meaning to thread jack here, but I've got a similar questions. I just purchased a Dell MD3200i with 12 x 600GB SAS 15K drives. I currently have two Dell R710's running ESXi 4.1 on them and will connect to the MD3200i via iSCSI. At some point I may grow to 4 ESXi hosts, but for now I'm connecting two. How many disk groups should I create? The setup wizard in the Dell utility recommended 3 disk groups with 4 disks in each. Any thoughts or recommendations?
This setup is for a test/training lab, so the I/O needs are not great. I was thinking about doing (4) 3-Disk RAID 5 groups with no hotspare. Like I said, this is a lab so downtime is acceptable. This setup would essentially give me a disk group for each host and the virtual disk for each group would be ~1.2TB. Any thoughts on doing it this way? This is my first time setting up a SAN from the ground up. Thanks for the help!
I would recommend against a single diskgroup for sure.
However, depending on your space, IO and read/write requirements, you could go with something like a 6-disk raid 10 and then a 5-disk raid 5 and 1 hotspare. The 5-disk raid 5 will have to be split into at least 2 virtual disks due to VMware's "2048GB minus 512 byte" maximum disk size limit.
With 6 disks, a single diskgroup is ok. But the original poster has 8 drives. When using raid 5, 2 smaller raid 5's can be better than 1 large raid 5 when using highly random IO (e.g. running multiple VMs).
Raid 10 is less impacted by this and 6 drives is small enough anyway.
Dev Mgr
6 Operator
•
9308 Posts
•
3 Points
506
0
Posted April 16th, 2011 23:00
ESXi's maximum LUN size is "2048GB minus 512 bytes", so your initial thoughts wouldn't work.
NLSAS is a glorified SATA drive; it has the same performance as a SATA drive, but it just happens to have a SAS PCB on the side (maybe a bit better testing and more reliable firmware, but the physical harddrive is the same as a SATA drive). It makes no sense to put your performance stuff on your slowest drives.
Last I checked VMware hasn't validated direct-connect iSCSI setups (only direct connect fiber channel or SAS), so for support reasons I would avoid that, but in theory if properly planned, it may work.
If you have the budget for an MD3220i with 8 x 300GB drives, 3 servers and a VMware Essentials plus license, I'd just do the setup right and get a set of decent iSCSI switches. If you want to stick with Dell, a pair of 5424's are more than enough (the 2800's aren't the best choice, the 5500 is overpriced right now, and the 6200, 7000 and 8000-series are way more than you need.
I'd also look into adding a dual port or quad port NIC to each server. The reason is that once you have shared storage, it would make sense to set up/enable vMotion. This requires a dedicated Gbit NIC (I usually use the same switches I use for iSCSI, but just create a separate VLAN on them for vMotion). So, with 4 NICs, you'd have 2 for iSCSI, 1 for vMotion and only 1 left for management and all your VMs to connect to your LAN... which lacks the redundancy that a setup like this deserves. If you add a dual port, I'd take 3 ports for management and VM LAN access (on vSwitch0 usually). If you add a quad port, I'd use 4 NICs for iSCSI, 1 NIC for vMotion and 3 NICs for management and VM LAN access.