EMS Alpha Build - Full Case
EMS Alpha Build - Just Drives
Server Side Drive Listing
Drive List - Client Side via Share
Power Usage on Initial Startup
(Startup/Surge Mode)
Power Usage - Standard "Run Mode"
(Post-Surge)
Power Usage with Drive Power Management Enabled
And finally, this is totally unrelated to the EMS, but is a picture of the "hard drive wall" which is just kind of cool! If you get really close you can see the 1s and 0s!!!
Posts
Do the PMs supply the power to the drives too?
ReplyDeleteGarrie
>> Do the PMs supply the power to the drives too?
ReplyDeleteThe Port Multipliers have two male molex style 4 pin connectors that supply power to the drives. They are underneath the PM board and you have to use 90 degree connectors with them as the standard ones won't fit.
One of the big mistakes that I did on the first round was to use a y split type cable for the PMs. In my case, the y split type power cable was too thin and will cause some problems when the drives spin out - not enough power.
What I did to get around this issue was purchase some larger cable - 16AWG I believe, run it to each PM and then split into the smaller Y Power Cable.
This effectively gave me a "larger power bar" - a 16AWG one with 2 - 20AWG taps versus a 20AWG one with 2 - 20AWG taps.
Don
Would be nice to be able to do a delayed power-up for the hard drives... Would allow us to use an even "lower-spec" PSU.
ReplyDeleteWith Drive power management the disks stop spinning when not accessed for some time, right?
And those 200W are measured will all the disk not spinning in "idle mode"?
The 200W is with just one active drive in the box.
ReplyDeleteNot to beat a dead horse - too late - that is one of the reasons I don't want a RAID setup as it you end up with a lot more spinning disks, vibration and so on which in a home environment probably isn't needed ???
I looked for a delayed spin up option and couldn't find one with the PM backplanes and/or SilImage Based Cards. I tried setting some of the drives to startup in 'standby' mode, via a jumper on them, but then the card never sees them.
I thought it would be a neat project to make a power board with a microcontroller that staggered the spin up and maybe had a USB port that would allow from software to shutdown individual PM backplanes. Too many projects, too little time! :)
Don
>>I don't want a RAID setup as it you end up with a lot more spinning disks..<<
ReplyDeleteSo if it runs in a RAID or JBOD, it will use more power because it has to spin all the drives up?
i.e. If its a 5 drive JBOD, it will "activate" all 5 drives even if the data is only on one?
And when you use all seperate drives, it only "activates" the one drive leaving all the others "spun down/powersave"?
Also, when using mount points, can you still spin drives down when they arent in use?
Sorry for so many questions. Just trying to get my "plan" before I buy everything.
Garrie
Garrie -
ReplyDeleteYes, if you have a standard RAID of X drives, all X drives will be active/spinning - there is no standard setup that I'm aware of that will power down individual drives in an active/ON RAID.
What I'm not sure of is if there are no locks or activity on files within that RAID is if the power management will allow the system to put all the drives to sleep. I'm sure if Windows doesn't, that you could force the issue with Linux though.
As for RAID off shoots like JBOD, I'm just not sure - you'd have to experiment...
When you use individual drives like I have, any disk without file activity will be powered down.
So for instance, right now, all 42 drives on my server are powered down/not spinning as there is no activity. If I play a movie, the drive that contains the movie will spin up. If I pause the movie, after 10 minutes (my Power Management Setting is currently 10 minutes) that drive spins down again. When I hit play, I have a 5 second pause before the movie restarts as I have to wait for the drive on the server to spin back up.
Mount points work just fine with power management. Just as you would expect for individual drives that are assigned drive letters.
For what it is worth, I've tried both camps. First was RAID as I was afraid to lose data. Frankly got tired of the heat and rebuilding a large RAID 5 array is no fun.
I think the single drive method is best - at least for me. I'd really like to see some real world experience with the JBOD side if anyone is willing. I'm pretty set with my current setup as it is working and the family gets cranky when I go into test mode :)
Another thing to think about, is that if you have a friend that is building the same sort of setup, you can use each other as your backup - that's a win win or a few levels ;)
Hope this helps!
Don
Have you considered using Stablebit DrivePool? I'd be interested if you have implemented any drive pooling options.
DeleteThis comment has been removed by the author.
DeleteThanks for the suggestion. Looks impressive and I particularly like the approach of still having standard access to the files (via a hidden folder on each drive) without needing specialized software in case something goes wrong...
DeleteHere's the site/link if you want to check it out too: https://stablebit.com/DrivePool
DeleteYes, its great software. I'm currently using it on my media server. I have ~22Tb, several "important" folders I keep redundant others not. Its really great software because if you don't care about redundancy you don't have to enable it for that particular folder.
DeleteIndeed, it's a fine balance between simplicity/reliability/efficiency/power economy/etc. :)
ReplyDeleteBut it's also the proof that we desperately need more appropriate File systems take can do all that work for us, and present us just with simple options.
(I'm no expert on file systems, but when I read a bit about ZFS it seemed like it is everything we could ever want. Guess we'll know in a couple more years if it ever gets to be implemented in a more "public" way. :)
Apple just ditched ZFS.
ReplyDeleteWhat are you using the Sata Controller on the system?
ReplyDeleteNever mind on last comment. need to scroll down on main page a little bit.
ReplyDeleteMy first thought would be noise, especially for a media server. I looked around (honestly not extremely in depth) on your posts that has no mention of this. Any ideas on decibel, or compare it to something common, id imagine its not a server rack loud, but damn near close enough.
ReplyDeleteBut sir, I do commend you on this project.
@slipvybe
ReplyDeleteWe already talke a bit about that in the comments somewhere - regarding the possibility of using low-noise fans.
If you're not filling the case entirely with drives, that would be an option - otherwise, it's best to keep a larger air flow to keep the drives cool.
i'm following this blog with great interest now that i've found it...i've been seriously considering something like this for a while and am glad to see a simple version out there ready to go.
ReplyDelete@ garrie
>>i.e. If its a 5 drive JBOD, it will "activate" all 5 drives even if the data is only on one?
the reason that the entire array spins up is that data striped across a RAID5 array isn broken up at file level, but at what ever block size the stripe was written across the disks at. these blocks sizes are usually pretty small for general file use, something like 64-128K...so for streaming a movie, or even an MP3, the stripe data has to be read off several disks in order to put the whole file back together. having to use more disks means you can take advantage of more spindles to answer the read for higher performance, but at a cost of power and heat.
this is even further complicated when you're writing data and all the disks with the new strip data (even those that the data may not have been written to if your array controller is smart enough) have to be read in order to recalculate the parity data that give RAID5 its fault tolerance.
the case for using the disks all separate, especially where large files/amounts of data can be efficiently streamed off each as needed is quite clear.
for those (like myself) that would like the best of both worlds...the automated backup of copying file from one disk to a secondary would be an easy choice without adding complexity. you could use mirrored pairs, or something similar, but that would just needlessly complicate something that in the end, is supposed to be very simple.
any backup script would have to be smart enough to only copy those files that have changed...with 90TB in total, copying 45TB to the other half would take forever without intelligently managing th e files that actually need to be backed up...
anyway...sorry this is so long, but you guys got me thinking...and i'm sure you all know what happens you get a nerd thinking ... ;-)
Hi Don,
ReplyDeleteI stumbled on your blog and would like to commend you on a job well done.
I thought my 27tb using direct attached storage was extreme :).
I am not sure if you have had a chance to checkout FlexRaid, its what I decided to use as a solution to add some redundency.
It eliminates a lot of the issues you brought up with trandional Hardware raid setups.
It uses a snapshot vs active and I have it sheduled to sync once a week when Im sleeping.
AWESOME!!!
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteThis comment has been removed by a blog administrator.
ReplyDeleteThis comment has been removed by the author.
ReplyDelete