JBOD vs RAID 5

[turak]

I've decided to upgrade my SageTV server and ordered the following:

Intel Core i7 920
ASUS P6T Deluxe V2
Corsair Domintator 3 x 2GB DDR3 1600
4 x Samsung Spinpoint F1 7200 RPM SATA 3.0Gb/s 1TB


I plan on using some of my existing HD's for the OS partition and dedicate the Spinpoint's to the SageTV recording. For tuner's I'm using two HD-PVR's (with AT&T Uverse) and a HD Homerun with both tuners recording OTA HD.

I am currently using RAID5 for recording and it seem's pretty stressed when I'm recording 4 HD streams, watching 1 HD program and comskip is running. How smart is SageTV when it comes to choosing which disk to write to when you have multiple storage folders configured? Will it try to spread the IO around or does it just go by which has the most space available.

I'm looking for performance not redundancy. I am wondering which would be the better way to go when the new parts arrive. Should I setup a separate recording folder on each disk or one RAID5 volume? Maybe I should just go with a 4TB RAID 0 stripe.

EDIT:
I would be using the onboard Intel SATA raid controller

[Djc208]

I think they were working on improving the recording intelligence but essentially Sage will record to whichever drive has the most available space. Which depending on how you load the drives from your existing system should mean that Sage will hop between drives as it records.

Do you need the redundancy, since that's the main advantage of RAID 5. I don't know if JBOD will help over just letting Sage see all four drives. To the user it would be invisible anyway. The RAID 0 stripe may help with throughput but also increases the risk of data loss. Plus most reviews don't show much improvement.

Personally I'd either just leave the drives alone, or consider running WHS and pooling the drives.

[turak]

My bad. When I said JBOD I meant just letting Sage see all of the disks. I didn't mean disk spanning.

[Lucas]

Originally Posted by turak

My bad. When I said JBOD I meant just letting Sage see all of the disks. I didn't mean disk spanning.

I think for recording it's best to let sageTV see all the drives.

You can also assign each of your capture devices to a specific disk.

This way there are no issues in recording multiple streams to a single drive which sometimes leads to problems.

I 've got 2 recording drives and it works quite well.

If you need to also have some recorded show security, you can also backup your recording drive contents to another occasionally or get WHS which does it through pooling and folder duplication.

[panteragstk]

I was under the impression that raid 0 used both drives bandwidth so theoretically it could double the bandwidth. 4 drives in raid 0 should be even faster. This is theoretical of course. When you factor read/write into the mix it gets odd. When reading only performance is great. Same with writing only. When you are doing both at the same time performance is pretty much crippled so it doesn't matter how much available bandwidth you have.

Here is a really good article on the subject. AnandTech SSD, SAS, SATA comparison.

[michelkenny]

The RAID 5 implementation on the Intel onboard controller is software based, which is REALLY slow. In order to have a fast RAID 5 array, you need to buy a controller with a hardware XOR chip, which usually runs at $150-$200.

[babgvant]

Stay away from RAID unless you really need the performance.

0 is faster in most scenarios, but if you lose a disc all the data is gone
5 fixes that using parity, but the implementation is controller specific so if the controller dies you need to replace it with the same kind or your data is gone (if you decide to go this way, get a discrete card)

I'd use JBOD; you'd have to need some serious I/O bandwidth to overrun a modern HD's capabilities. It maximizes storage space and limits risk (only the drive that fails data will go away). If Sage is good about utlizing all the drives in your collection, you'd have plenty of I/O headroom (recording 1 writes to disc a, while recording 2 writes to disc b).

[stanger89]

Originally Posted by panteragstk

I was under the impression that raid 0 used both drives bandwidth so theoretically it could double the bandwidth.

Sequential bandwidth, but that's not the limiting factor for a DVR application. DVR applications are bound by latency, this is why 64k clusters are recommended, it allows more data to be read before having to seek to new data.

RAID-0 doesn't do anything for latency, as that's determined almost exclusively by rotational speed.

[turak]

Originally Posted by babgvant

If Sage is good about utlizing all the drives in your collection, you'd have plenty of I/O headroom (recording 1 writes to disc a, while recording 2 writes to disc b).

This was the point of my question. I understand the mechanics of the various RAID types. What I was/am unsure about is how intelligent Sage is about handling multiple disks. Let's say we have 4 disks called A, B, C and D. Drive A has 800GBfree and the rest have 750GB. Let's assume that SageTV just uses whatever has the most free space. I have 4 shows scheduled to record at 8:00 PM. Since all four shows start at the same time and Drive A has the most free space at 8:00 PM, then I would think all four recordings would go to the same disk.

If this example is how Sage makes use of available disks then I would probably get better performance with RAID since it would be spreading the IO across multiple spindles. If instead it splits active recordings among the available disks then I would get better performance with just a bunch of disks.

Someone mentioned assigning a capture device to a disk. I did not know that you could do this. Is this a merit system or will it ONLY use this disk?

[babgvant]

Originally Posted by turak

This was the point of my question. I understand the mechanics of the various RAID types. What I was/am unsure about is how intelligent Sage is about handling multiple disks. Let's say we have 4 disks called A, B, C and D. Drive A has 800GBfree and the rest have 750GB. Let's assume that SageTV just uses whatever has the most free space. I have 4 shows scheduled to record at 8:00 PM. Since all four shows start at the same time and Drive A has the most free space at 8:00 PM, then I would think all four recordings would go to the same disk.

If this example is how Sage makes use of available disks then I would probably get better performance for RAID. If instead it splits active recordings among the available disks then I would get better performance with just a bunch of disks.

If it just picks the one with the most space it should work out the way you want over time. You could proactively balance recordings as well.

[PLUCKYHD]

Originally Posted by babgvant

If it just picks the one with the most space it should work out the way you want over time. You could proactively balance recordings as well.

I recomend NO raid or WHS myself.

[turak]

Originally Posted by babgvant

If it just picks the one with the most space it should work out the way you want over time. You could proactively balance recordings as well.

How so? If four recordings start at the same time and if it's using the drive that has the most free space at the start of the recording, then all four recordings will always go to the same disk. If it's going by how much free space will be available after the recording is finished then it would probably end up balancing them across disks.

[Skirge01]

Like others have said, unless you're sure you are going to have an I/O issue with your disk subsystem, then I'd avoid RAID. I run WHS myself using a mix of 5400 rpm WD GP (they're actually around 6000 rpm) drives and 7200 rpm Samsung F1 drives (10 drives in total so far) which are all part of the WHS drive pool, which is, technically, a software RAID. I have never had an issue recording 4 HD shows, while watching a 5th. Admittedly, I was originally planning to use RAID 5 or even RAID 6 (at a cost of >$1000 for the controller card alone), but after much research, I concluded the following reasons were enough to stay away from it...

Keep in mind that you increase your risk of failure by going to RAID. RAID 0 is the worst, since you lose everything when a single drive failes. RAID 5 protects you from a single drive failure, but anyone who has set up an array and attempted to recover after a failed drive will tell you that you're not guaranteed to be able to rebuild the array after a failure. Additionally, during the rebuild, your IO is significantly degraded and another drive failure would also cause you to lose all the data. Finally, adding storage space (capacity expansion) can be problematic (doesn't always work) and if the RAID controller fails, you may not be able to just pop in a new controller, unless it's the exact same model.

Back to my original point... try it without RAID first. If you do run into IO issues, THEN start considering some options such as WHS, NAS, etc. Don't go overboard until you know there's an actual issue to address.

[babgvant]

Originally Posted by turak

How so? If four recordings start at the same time and if it's using the drive that has the most free space at the start of the recording, then all four recordings will always go to the same disk. If it's going by how much free space will be available after the recording is finished then it would probably end up balancing them across disks.

I looked in the manual; Sage load balances using drive space and expected recording size, so over time it should balance recordings across the drives as they are used.

In your scenario; depending on the expected size, 2 or more recordings may get allocated to one disc. But if you use the same drive size and start from 0 (no recordings) with all the drives, that scenario (a large gap in free space) should only happen when the shows you've deleted are disproportionately on one drive.

[babgvant]

Originally Posted by Skirge01

WHS drive pool, which is, technically, a software RAID.

unless something has changed since release, WHS isn't any kind of RAID; it's a fancy version of JBOD where all the writes are taken on the primary drive.

[Skirge01]

Originally Posted by babgvant

unless something has changed since release, WHS isn't any kind of RAID; it's a fancy version of JBOD where all the writes are taken on the primary drive.

Okay, okay. You're right. I was trying to keep it simple for people. Drive Extender (the "drive pool" of WHS) is a pretty complex method of sharing the disk space between multiple drives, but only requiring the user to know where the share is. For example, everyone knows c:, d:, e:, etc. On a network, you have \\server\share1, \\server\share2, etc. With DE, you could have a single share (i.e. share0), which contains links to 20 different physical hard drives. When you put a file into the share0 folder, DE will determine which physical disk it needs to go onto and will balance the data so no drive is being utilized a near capacity while another sits completely empty. You simply see a shared folder with the pooled space from all the drives in the drive pool, hence the name.

EDIT: Although, if you utilize WHS' duplication feature, that's essentially RAID 1, but that wasn't a feature I was originally talking about so babgvant is still correct.

EDIT #2: Also, all the writes are no longer taken on the D: drive (secondary partition of the system drive aka the "landing zone"). MS got rid of the landing zone a while ago.

[turak]

Originally Posted by babgvant

I looked in the manual; Sage load balances using drive space and expected recording size, so over time it should balance recordings across the drives as they are used.

This is what I was looking for. I guess next time I should RTM.

[GTwannabe]

RAID 5 has rather pitiful write speeds because the parity stripe has to be calculated, written, and verified, so JBOD/RAID0/RAID1/etc will be much faster for Sage use.

[lobosrul]

I use a Thecus n5200 in raid-5 (5x750GB) for archiving. It ran off parity after I had a disk go bad. So it really does work. Its not good for recording though. Just two HDTV streams caused stuttering.

[stanger89]

Originally Posted by GTwannabe

RAID 5 has rather pitiful write speeds because the parity stripe has to be calculated, written, and verified, so JBOD/RAID0/RAID1/etc will be much faster for Sage use.

That all depends on the implementation. A good RAID controller can do RAID-5 with little/no throughput hit.