VMware Cloud Community

ESX 3.0.3/3.5 SAN Upgrade Question / Advice request

I have 2 ESX 3.0.3 servers right now both connected to an HP MSA 1510i iSCSI SAN. Originally this project was used to virtualize some older hardware that needed to be unplugged. I'm trying to plan the best upgrade path now, to expand and use the 2 exiting ESX servers to virtualize most of not all of the data center. This will include Exchange, a File server, etc. I will probably also host a (physical) SQL server on the same storage. The 1510i will be used to host shared files, and backup needs. I have about 100 users on site.

I don't get terrific performance off of the 1510i, obviously. I have 4 7.2k SATA discs in a RAID 5 configuration. I'm using single path, and single controller. No HBAs. I've been running some IOMeter traces, comparing discs hosted on the MSA vs. discs hosted on local storage, and depending on other activity on the VMs, I'm getting 3 to 6 times higher transaction time with the MSA.

I'm looking to upgrade the storage infrastructure to accommodate this plan. With my budget for this, I'm looking at an AX4-5 or SMS100 class SAN.

My question is, do I stay with iSCSI? Or do I move to 4Gb FC? The iSCSI would involve buying a couple Cisco 2960 Gigabit Switches for redundant fabric. FC, I'm probably looking at Brocade 300 Fabric Switches, and then HBAs. The cost of the FC is going to be possibly 20% higher then sticking with the iSCSI, which I could work into the budget if needed.

Does my situation warrant making the jump into FC? Or am I overengineering this scenario?

If I go FC, is the extra cost of a Brocade 5100 vs. the 300 going to do anything for me? I'll probably never need more than 16 ports, and I can only afford 1 5100, whereas I'll have redundant fabric with the 300s. I'm a little worried becuase the 300 isn't lited under Hitachi's storage documents for VMware (while it is listed for MS Windows).

If I stay with iSCSI, should I get HBAs? Do I need to get HBAs in order to copy the VMs from the 1510i to the new SAN? (Because VMware software initiator only sees 1 target?)

Thanks for your help,


0 Kudos
1 Reply

Your options will most likely depend on several factors but for latency sensitive applications like Exchange and SQL the ability to control multi-pathing with FC from the ESX hosts is usually a good idea. I would separate the data into raid groups based on the IO profile (sequential writes vs random reads.) I would also take this opportunity to monitor the existing IO and throughput profiles of the physical Exchange and SQL servers. File servers should not be a problem if designed correctly.

It all depends on what the current environment is doing and your planned rate of growth. You should also account for the impact of a host failure and all your IO coming from a single host.