I manage several Dell/Equallogic MD1000 storage arrays. They require careful understanding of their limited redundancy features.
I find that it helps to remember this: the MD1000 is a JBOD. Period. Any data redundancy actually depends on your RAID controller.
Although most components are modular, that only helps with field repair, not availability:
| Device | Component | Feature | Note |
|---|---|---|---|
| MD1000 | Disks | Hotswappable | Yes, but depends on your RAID controller. You should offline the disk first. Note that rebuilds involving large volumes will take weeks to complete, so use 73GB disks if that’s a problem. |
| Redundant | Yes, depends on your RAID controller. | ||
| Cables | Hotswappable | No, requires host and array power down first. | |
| Redundant | No, MD1000 alone does not support multipath like the MD3000. | ||
| Power Supply | Hotswappable | Yes | |
| Redundant | Only for 5 minutes since 3/4 fans are required to cool. | ||
| Fans | Hotswappable | Yes, as part of Power Supply module. | |
| Redundant | Yes, 1 spare with both power supplies working. | ||
| EMM | Hotswappable | MD1000 alone: Not really for EMM0, since removing EMM0 loses disk communication, and failback is not automatic. EMM1 can be hot swappable if enclosure is in unified mode, thus not used. Also, the host must be rebooted. Connected to MD3000: supposed to be hotswappable. | |
| Redundant | MD1000 alone: No. There can be 2 EMMs installed, but if EMM0 fails, then some disk communications will be lost, whether unified or split mode. Also, no automatic failback. EMMs provide redundancy for enclosure control functions only. Connected to MD3000: supposed to be fully redundant. | ||
| Front Panel | Hotswappable | No, requires power down and enclosure disassembly to replace. | |
| Redundant | No, only 1. | ||
| Firmware Updates without Reboot | Server | Maybe with latest firmware, but a reboot is recommended (common sense), especially with multiple arrays. | |
| MD1000 | Maybe with latest firmware and Perc 6/E. See R216024. A reboot is recommended (common sense), especially with multiple arrays. | ||
| Clustering | MD1000: No. MD3000: Yes, should work if you’re lucky. |
Notes:
- A backup is recommended and needed before any configuration changes are made in hardware or software.
- Adding a MD1000 to an existing daisychain requires latest firmware to be loaded first. Mixing firmware is undefined. Also a reboot has been required in the past.
- Adding a MD1000 to MD3000 daisychain requires reformatting of MD1000 disks.
- Adding a MD1000 to a MD1000 daisychain renumbers the enclosure IDs.
- All MD1000s must finish startup before MD3000s, and before host servers, or disks will be “missing” or “foreign”. So configure servers to “off” after power failure.
- In fact, Perc controllers can “forget” their volume configuration after reboot, so put everything on a UPS and never reboot.
- MD3000 with MD1000s daisy-chained from 2 HBAs supports redundant cables.
- Do not read log if a rebuild is in progress.
- Snapshots can be done with linux and any filesystem that supports it, such as ext3 with LVM.
So what are you paying for? Basically, a well-manufactured, well-tested, non-HA, field-serviceable DAS device with good Linux and Windows support and reasonable performance without licensing encumbrances (well, Dell has starting to restrict disk replacements.) Most people who buy other no-name JBOD devices that I’ve talked to never get something that works right, though SuperMicro multi-disk servers seem to be popular.
Choosing the iSCSI versions could avoid a lot of the cabling and reboot order issues.
Please add a comment with your suggestion or tip about managing the MD1000 or MD3000, or recommendations for affordable, more-available arrays.
Dell Powervault MD1000 Manuals
MD3000 and MD3000i – Generation 2 Firmware Update (2008-12-21)
Dell Support Forums: MD1000 Redundant EMM?
bladewatch.com: Dell Server Firmware
dell.com: PERC 5/E Fault Tolerance Features
ftp.dell.com: Dell PERC 6/i Integrated Firmware Update 6.2.0-0013 – R216024
INetU Labs takes on the Dell MD3000i: Is it an Enterprise-capable workgroup SAN?
IDC numbers show Dell server storage booming
theregister.co.uk: Dell servers block un-Dell HDDs
dell.com: Third-party drives not permitted on Gen 11 servers
dell.com: Why Customers Should Insist on DELL™ Hard Drives for Enterprise Systems
cyberciti.biz: RAID 5 vs RAID 10: Recommended RAID For Safety and Performance
Thought you might want to read this given our past conversations:
http://makarevitch.org/rant/raid/
a good read, lots of numbers, boils down to use RAID 10, RAID controllers are expensive chassis ornaments, and potential bottle necks.
Avoid RAID 5. ( & 6 their parity xsum is too costly for rand IO, such as internet file servers, DBs etc. )
Not mentioned in the rant but read elsewhere, putting the filesystem journal on an SSD could help IOPS along nicely as well.
-R