7

I support a non profit. They got an ML350 G5 donated to them and I immediately put it to good use for them :) It came with four HP 300GB 15K SAS drives - and of course, they have all now failed. I have been replacing the drives and they are all 6Gb/s drives, but the HP controller in the ML350 is a 3Gb/s controller. I have a nice 8 port LSI MegaRAID SAS controller sitting here with battery backup and everything - can I just swap out the controllers and expect everything to work?

I haven't found much about the differences between 3 and 6 Gb/s SAS interface from a physical layer. As near as I can tell, it should work just fine. I'm just curious if anyone has any experience doing this.

I'm not hung up on the drives being hot swappable, so if not I may just see if I can physically alter the case, toss the HP drive cage and stuff a different cage or mounting system in there to accommodate the better controller with the newer drives.

Ugh - now I remember why I prefer white box servers for one-off or small business vs. the "big boys" like HP, Dell, etc...

And anyone have a good source for rails for this thing? The HP prices are ridiculous - I could build another server for what they want for rails!

ewwhite
  • 194,921
  • 91
  • 434
  • 799
EricEE
  • 73
  • 6
  • Google the part number of the rails, (or even just the server model number and "rail kit") and you should be able to find plenty of sources for used rail kits. As to white-box servers vs OEMs, you're entitled to your opinion, but... well, white box servers don't come with a warranty or service agreement, which is something that's nice to have when anything breaks at 3 in the morning. – HopelessN00b Oct 09 '12 at 04:25

5 Answers5

7

That's a lot to address. I'd keep the existing Smart Array p400 controller in place for now, provided it has a working battery-backed cache unit. Otherwise, your LSI will work, but it's not ideal.

The on-disk RAID formats are different, so this means a full rebuild. You won't be able to preserve your data if you do what you're proposing.

As for 3G versus 6G, it does not make a difference unless you're dealing with SAS expanders or solid-state drives. You won't see any difference in performance in your case.

I tend to think that HP controllers run better in their servers. I'm also not a fan of LSI's management software (MegaCLI is awful). Using LSI HBA cards for other solutions makes sense, though. Your best bet in this case is to make sure you have an HP P400 controller with BBWC cache (here's why). If you don't have the cache, your path of least resistance is to obtain one.

That way, you can be sure that your backplane, alerts, LED indicators and integration with system monitoring all work properly.

As for rails, check: http://www.racksolutions.com/hp/racking-hp/ml350

ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • Thanks - I didn't think it was a big deal. since I need to redo the box anyway and I have a couple of new drives to add to the array, I'm going to try out the LSI. I think just to be safe I'll get a disk perf/stress test utility and let it "burn in" for a bit, but I don't see why the HP backplane should cause issues if I swap the controller and the cables too. Thanks everyone! – EricEE Oct 08 '12 at 16:51
  • I'm also looking forward to upgrading to RAID 10. I wish the drive cage held 8 drives instead of just six. I've been tempted to see if I can find either two four or one 8 bay cages that would physically fit in place of the HP cage. LSI has some great caching software that has made a HUGE impact on other servers I have set up, but I think I'm going to loose use of two of the 8 ports because of the HP backplane. I need to get back in the server and review the physical layout again... – EricEE Oct 08 '12 at 16:56
  • Thanks for the racksolutions link - I forgot about them! They have a perfect looking 2 rail flush mount kit. If I scrounge up some extra funds I might just splurge! – EricEE Oct 08 '12 at 17:07
  • @EricEE See my updated post. – ewwhite Oct 08 '12 at 17:11
  • Thanks eewhite. Yup, I think MegaCLI is ugly - but their controllers are top notch, and CacheCade seriously kicks butt! I don't care about the HP integrated management - more of a PITA than not, if you ask me. I've already got a VM set up for syslog, CACTI and a few other management tools - all with email alerts. Covers my entire infrastructure, not just the HP specific stuff. I tend to find the vendor-specific solutions deceptive. At first they seem like a good deal because they are "free", but very quickly you find yourself locked into them to keep that "ease of use". – EricEE Oct 08 '12 at 17:17
  • @EricEE In larger or more structured environments, these things do count. I really wouldn't overlook the value of the right monitoring tool/agents. – ewwhite Oct 08 '12 at 17:30
  • And as for your megaraid link, I'm not interested in JBOD mode so no worries there - but it's good to know. I've played with ZFS off and on, but unfortunately I've found ZFS to have more downs than up. Just search on long ZFS recovery or failures. Yup, I back my stuff up (neither RAID nor ZFS is a backup solution) - but the reason I want fault-tolerant hardware/software designs is if I loose one drive I don't *have* to reload everything. So for right now I'm reaping the benefit of cheap hard drive prices and sticking with RAID 10. Not the most efficient but fast and easy to recover! – EricEE Oct 08 '12 at 17:31
  • "In larger or more structured environments, these things do count." Yup. Just not in mine :-) "I really wouldn't overlook the value of the right monitoring tool/agents." I don't. That's why I have them - it's just I don't have to rely on HP's (I'll use 'em if they are there, but I'm not going to get hung up on 'em). – EricEE Oct 08 '12 at 17:31
3

Ugh - now I remember why I prefer white box servers for one-off or small business vs. the "big boys" like HP, Dell, etc...

Building your own is almost never cost effective. By the time you factor in your time spent building the machine, figuring out what is supported, warranty, future support, downtime if something goes wrong, downtime hassling with vendor warranties (or keeping spare parts)... TCO is not just CapEx.

And anyone have a good source for rails for this thing? The HP prices are ridiculous - I could build another server for what they want for rails!

Try eBay and other outlets. You can usually find them cheap. Much less than HP would charge directly.

I haven't found much about the differences between 3 and 6 Gb/s SAS interface from a physical layer.

The only difference is the speed of the data; and physically speaking the cables are a higher quality to support the higher speeds. The connectors used in the original spec call for 10GHz minimum.

I have a nice 8 port LSI MegaRAID SAS controller sitting here with battery backup and everything - can I just swap out the controllers and expect everything to work?

That would depend on what controller you're using now. If you're using the SmartArray controller, no. You would need to rebuild the array.

Unless you're using new 15k drives or SSDs you will not be saturating a 3Gb bus, so you wouldn't see any performance gains from swapping to 6Gb. Also the LSI card would not be warrantied by HP.

four HP 300GB 15K SAS drives - and of course, they have all now failed

What on Earth did you do to those drives? I've never seen drives fail that quickly. I've got dozens of such drives and only one has failed in 4 years.

Chris S
  • 77,337
  • 11
  • 120
  • 212
  • All the drives were probably original with the machine. Only two of the original five are totally dead, the rest are "pre failure" which pretty much means I'm not going to mess with 'em. Again, warranty is out the window at this point. And yes, the new drives are 15K. I didn't even think about it originally since I didn't pay attention to just how old this server is in the grand scheme of things. – EricEE Oct 08 '12 at 16:49
  • "TCO is not just CapEx." Depends on the environment and the requirements. Dell, HP and others will often limit your choices for what to do if/when a problem arises. And to ensure you have that warranty, you are definitely restricted on upgrades or expansion. At the low end of the spectrum, I can afford to have a complete second machine sitting there as a cold spare for the "convenience" of paying Dell/HP/etc. for their warranty support. – EricEE Oct 08 '12 at 17:22
  • As with any insurance policy (which is what a warranty is), it's about trade off's. In my situation we roll the dice and self insure - mainly because we have to. And we usually come out on top. If warranty's were such a good deal for the consumer, BestBuy and other retailers wouldn't be pushing them so hard! Same reason Dell upsells them on every other screen. As for time, as stated I'm a volunteer and while my time is still valuable, hard cash is even more valuable. It's part of my donation. Whitebox parts are also easier for others to support since there are no vendor restrictions. – EricEE Oct 08 '12 at 17:24
  • The warranties sold by BestBuy (where there's extremely little benefit, and very high relative cost) really aren't in the same class as a HP/Dell 13x9x4 Warranty. Lumping them together like that is a [composition fallacy](http://yourlogicalfallacyis.com/composition-division). While all business decisions should be about the trade-offs, you lead your reasoning to an outcome expressed as "we have to". Your client is apparently very tolerant of downtime, so your requirements are different than most businesses. This variation in business requirements is what makes the white boxes appealing. – Chris S Oct 08 '12 at 18:32
  • I work with some very large non-profit corporations. Their downtime requirements necessitate Dell/HP Servers and Warranties. It's not the fact that their non-profit. – Chris S Oct 08 '12 at 18:34
  • Well, not all non-profits (as well as business) are the same. We are not a large non-profit corporation. We are small and cash strapped, and make due with the resources we have. What doesn't get donated by others usually gets donated by me! I'm not pitching this as a solution for everyone - one size definitely doesn't fit all. And yes, the BestBuy comparison may not be directly applicable but it's close enough for my environment. – EricEE Oct 08 '12 at 19:17
  • As for "tolerance of downtime" - I don't know of anyone who likes downtime. We just can't afford to take the steps to ensure it doesn't happen by having redundant hardware clustered and under four hour maintenance windows, etc. For what we do, having a day or two of downtime isn't worth the costs to prevent it. Simple cost/benefit analysis. That doesn't make us any less legitimate than anyone else. If you can afford it, or you have requirements that justify the expense, then great! That's why those services are there. – EricEE Oct 08 '12 at 19:21
  • But even if we could afford it, is the best use of our resources? Could that money be spend where it might do greater good vs. reducing staff inconvenience? We don't have critical systems. If our servers go down, we can function without them for brief periods with relatively little interruption. Not everyone is the same. If their systems go down they are down and losing money (or putting service in jeopardy). That's not us, so hopefully that explains what some no doubt view as a "cavalier" attitude on my part. – EricEE Oct 08 '12 at 19:23
2

Please check for hard drive firmware upgrade as well as RAID controller firmware upgrade, there was a notification about false hard drive pre-failure depending on drive type. I have never seen that many drives in one server fail or pre-fail unless something else is an issue. We have drives running for 6 years without an issue. If you are supporting HP hardware please know that you can sign up for their alerts here https://h41183.www4.hp.com/user-prefs-alerts.php

Amir
  • 21
  • 1
1

You can remove the failed drives from the drive carriers and replace with a new drive. Several folks make rails/carriers and I have also bought dead drives just to get the drive carrier. Make sure they are enterprise class drives or you may have issues with the array reporting failed drives. WD RE drives are a good one we have used. There are others as well.

Swapping the controller may give issues as well so back everything up. Back it up no matter what!

Also consider if the 6Gdrives will gain anything for the client. Is drive I/O a bottleneck? If not the existing controller may be fine.

While building a server may cost less, what about support? If you get hit by a bus who do they call. With the mainstream guys, parts and servcie are a call away. I had a Dell drive fail and had a replacement in less that four hours even though we did not have 4 hour support and are not in a major center.

Dave M
  • 4,494
  • 21
  • 30
  • 30
  • Thanks for the responses. I know I have to rebuild the array - that's not an issue. The HP card has battery backup, so does the LSI. I want to rebuild the array and re-load the host OS - it's a HyperV host - my servers are all HyperV guests. And since I am going to rebuild the array anyway, my thinking was to replace the controller to get the best performance I possibly could. – EricEE Oct 08 '12 at 16:45
  • Oh, and as for TCO and support - they will do what they always do - get another volunteer :) This is a non profit. While name brand stuff may be a no brainer for business, it's just not an option for us most of the time. Everything is documented (my pet peeve) so anyone with half a brain coming in behind me can see what's up in 10 minutes tops. If it hadn't been donated there is no way they would ever have this class of a machine for a server. Thankfully someone tossed it to us instead of the dumpster and it's a great hyperv host! – EricEE Oct 08 '12 at 16:53
  • One final note - I'm rebuilding the HyperV host to upgrade to Server 2012 and replace the current host OS install that is based off the MultiPoint Server Premium 1+1 licensing - when you use MultiPoint Server for your host HyperV (it's allowed) you have to leave the MultiPoint stuff there (just unused). I don't like all that extra stuff in my hypervisor OS; I'm not scared of server core (and again, I document for those coming behind me) so it seems like the perfect time to spice things up. Thanks again for everyone's input! – EricEE Oct 08 '12 at 17:10
1

Regarding 3GB/s vs. 6GB/s, you shouldn't have any problems (from the drive's side of things) making the swap. I don't recommend replacing the drive cage and connecting the drives directly to the LSI controller, though. It's unlikely that the server's power supply has the right type of power connectors available to do that sort of thing. The server is pre-wired assuming that drives are going to be sourcing power from the backplane, and there are usually only one or two single-device power cables available (for use with optical drives). I've tried to do this before, and it ended up being more hassle than it was worth. Some server models may handle this better than others, so YMMV.

Regarding server rails, the HP prices don't seem bad. The HP website has a rack mount kit that works with a ML350 G5 and currently sells for $61.

bta
  • 536
  • 2
  • 8