RAID 0 basics: Optimal stripe size for best performance (for HDTV recordings)

[mkanet]

I know for single disks NTFS 64k clusters are recommended for sageTV recordings. I'm curious what's the stripe size thats recommended for Raid 0 (I'm going to use 2 sata disks). Raid 0 dedicated to sageTV TV recordings only. I'm not sure if it matters to say, but I have 32MB's of internal cache per HDD and an Intel ICH8. Obviously, I would like the setting for the best performance. If I was to use RAID 5, or anything which supports parity.. Id no-doubt get a dedicated higher end RAID controller ie a PERC 6, etc.

If there are any additional tips you think I coud benefit by, please let me know.

[sic0048]

I don't think you'll find any benefit to using RAID 0. Just how many streams of HD are you trying to record at once?

[mkanet]

I already have 2 identical disks Id like to use. Id like to get the best throughput while I can. So, if I tell you how many recordings Im going to record at once, you'll answer my question?

I guess a maximum of 10 recordings plus any playback streams on top of that.

Originally Posted by sic0048

I don't think you'll find any benefit to using RAID 0. Just how many streams of HD are you trying to record at once?

[sic0048]

I guess I should have looked at your signature

Are you seeing through-put issues with your current set up? Obviously RAID 0 will give more performance, I've just never seen anyone complain that a regular SATA setup was too slow for them for normal Sage use. But again, you aren't the average user either. That is one hell of a setup.

PS - I haven't checked, but does the Intel ICH8 chip have a stripe size limit? I know some hit a limit.

I think 64kb or 128kb would be a nice middle ground personally.

[GKusnick]

I don't have an answer on the stripe size question. But if I'm doing the math right, 10 HD streams at ~8 GB/hour works out to ~20-25 MB/sec, which is still much less than the rated throughput of single SATA drives, which (as I recall) is typically upwards of 60 MB/sec. So I'm somewhat skeptical that striping is really going to buy you any measurable improvement over, say, using forced_video_storage_path_prefix to direct half your tuners to one drive and half to the other. And of course the big downside to striping is that if either drive dies, you lose all your recordings, not just half of them. Seems like the risk/benefit ratio doesn't really add up.

[mkanet]

Well.. hopefully, someone will know the answer soon before I'm ready to build the RAID (hopefully tomorrow).

It's pretty easy to overlook the advantages of RAID0 (and other RAID levels) since the obvious primary advantage is read/write throughput. For me, its a combination of needing more throughput and also reducing wear and tear of a single disk. I think people would be surprised how quickly I wear out a single disk banging away recordings 24/7on many tuners.

When a single disk is busy recording 4 HD recordings at a time. If I try to do FF/REW functions on a 5th or 6th recording, I notice noticable performance degredation/lag: brief pixelization of video for the first split second and takes a split second longer to ramp up the throughput necessary for FF/REW. After about 6-7 simultaneous recordings, the disk start pooping out with audio/video stuttering in the recording. So, whatever the theoretical limit is for a single sata disk, considering the mechanic's of the SATA disk and SATA protocal overhead, the real-life read/write throughput adds up to roughly 4-5 recordings without having any kind of lag during FF/REW.

I personally dont like the idea of having multiple logical disks used with forced_video_storage_path_prefix; as I prefer to see all my recordings in one place.

I dont really mind if one of my disks goes out and lose my recordings since I backup all the recordings/movies I really care about.

Originally Posted by GKusnick

I don't have an answer on the stripe size question. But if I'm doing the math right, 10 HD streams at ~8 GB/hour works out to ~20-25 MB/sec, which is still much less than the rated throughput of single SATA drives, which (as I recall) is typically upwards of 60 MB/sec. So I'm somewhat skeptical that striping is really going to buy you any measurable improvement over, say, using forced_video_storage_path_prefix to direct half your tuners to one drive and half to the other. And of course the big downside to striping is that if either drive dies, you lose all your recordings, not just half of them. Seems like the risk/benefit ratio doesn't really add up.

[paulbeers]

What you are experiencing has nothing to do with thoroughput, it has to do with thrashing. Too many things happening on one hard disk. However, by going to Raid 0, this wouldn't it theoretically INCREASE. Why is that? Because right now in your scenario of 5-6 items (4 recordings and 2 playback), I bet at least one of those is happening on drive 2. So really you might have 4-5 items on one, and 1-2 on another. If you move to RAID0, all hard drives are being hit at all times! Since half of the show is on one, and half the show is on another, then you are thrashing both hard drives at the same time! Rather than just really hitting one. You would be better off with several smaller drives or setting your tuners to record to certain hard drives (I knwo you have already stated you don't want to do this).

Also, while I know you stated you don't care about losing your recordings, but it seems to me losing all recordings would be kind of annoying, rather than just losing half if a hard drive failed.

As a side note, when I have run RAID0 in the past, I found limited real world advantage to it. Nothing "felt" faster. Sure my benchmarks were better and load times with slightly better, but it wasn't worth the risk!

[gplasky]

What you may want to consider is using your RAID 0 strictly for the recording of shows and only that. Possibly monitor that with Dirmon2 and actually move the recordings to another drive for any viewing when done. Maybe also comskipping on those drives before moving them. This way your watching habits will never be impacted by your recording habits no matter how many streams.

Gerry

[mkanet]

You have it almost right, except, you would have half the amount of trashing per disk.

Originally Posted by paulbeers

What you are experiencing has nothing to do with thoroughput, it has to do with thrashing. Too many things happening on one hard disk. However, by going to Raid 0, this wouldn't it theoretically INCREASE. Why is that? Because right now in your scenario of 5-6 items (4 recordings and 2 playback), I bet at least one of those is happening on drive 2. So really you might have 4-5 items on one, and 1-2 on another. If you move to RAID0, all hard drives are being hit at all times! Since half of the show is on one, and half the show is on another, then you are thrashing both hard drives at the same time! Rather than just really hitting one. You would be better off with several smaller drives or setting your tuners to record to certain hard drives (I knwo you have already stated you don't want to do this).

Also, while I know you stated you don't care about losing your recordings, but it seems to me losing all recordings would be kind of annoying, rather than just losing half if a hard drive failed.

As a side note, when I have run RAID0 in the past, I found limited real world advantage to it. Nothing "felt" faster. Sure my benchmarks were better and load times with slightly better, but it wasn't worth the risk!

[mkanet]

Exactly...

Originally Posted by gplasky

What you may want to consider is using your RAID 0 strictly for the recording of shows and only that. Possibly monitor that with Dirmon2 and actually move the recordings to another drive for any viewing when done. Maybe also comskipping on those drives before moving them. This way your watching habits will never be impacted by your recording habits no matter how many streams.

Gerry

[mkanet]

Okay, after a lot of searches on google. It seems the best answer to my question is a 128kb stripe size for video editing/DVD authoring scratch disk. I think my controller supports up to 128kb stripe size anyway. Well see what happens when I try it today. Who knows, maybe this will be a big waste of time; but, at least I'll know for sure

[GKusnick]

Originally Posted by mkanet

You have it almost right, except, you would have half the amount of trashing per disk.

I don't see how that follows. Ignore recording for the moment. When playing back two streams from separate disks, seeking is minimal (assuming files are stored more or less contiguously) because each disk is reading from only a single file.

On the other hand, when playing back two streams from a RAID0 stripe set, both disks must seek continuously, because each supplies half the data for each stream, and must therefore shuttle back and forth between them.

When you add recording to the mix, things get a bit more complicated. But it's still true that with two separate disks, each disk has to do only enough seeking to accommodate half the total streams. With RAID0, each disk must do enough seeking to accommodate all the streams, since every stream touches every disk.

The case where RAID0 shines is when reading or writing a single sequential stream faster than a single disk's native throughput. In that case striping lets you queue data alternately to or from each disk and double the effective throughput. But that's not at all a typical usage pattern for SageTV, where single-stream bit rates are far below the disk's native speed, and multiple streams (especially simultaneous playback) generate additional seeking that kills throughput.

[Beefcake550]

From my experience, the answer to your question is either 64KB or 128KB stripes. I know you already figured this out, but reassurance is still helpful . The way to determine the answer for you is to run some disk benchamrks which simulate your environment the best.

Good Luck....

[MitchSchaft]

I'm with the other guys. I doubt you'll notice any difference. Not to mention the more parts you have the greater chance you have for something going wrong. But, it sounds like your mind is already made up so just go ahead and do it .

[paulbeers]

Originally Posted by GKusnick

I don't see how that follows. Ignore recording for the moment. When playing back two streams from separate disks, seeking is minimal (assuming files are stored more or less contiguously) because each disk is reading from only a single file.

On the other hand, when playing back two streams from a RAID0 stripe set, both disks must seek continuously, because each supplies half the data for each stream, and must therefore shuttle back and forth between them.

When you add recording to the mix, things get a bit more complicated. But it's still true that with two separate disks, each disk has to do only enough seeking to accommodate half the total streams. With RAID0, each disk must do enough seeking to accommodate all the streams, since every stream touches every disk.

The case where RAID0 shines is when reading or writing a single sequential stream faster than a single disk's native throughput. In that case striping lets you queue data alternately to or from each disk and double the effective throughput. But that's not at all a typical usage pattern for SageTV, where single-stream bit rates are far below the disk's native speed, and multiple streams (especially simultaneous playback) generate additional seeking that kills throughput.

Thank you Jerry for making my point more clear! Every drive has to supply every stream! Drives work best when they are supplying one large file because the heads just reading sequentially on the disk. When you are accessing two large files at the same time, the heads have to constantly move. So, the nature of Raid 0 is that it is NOT good for tons of non-sequential reads such as in this case where it will have to read 6 files all at the same time from 2 different disks. In essence, to play back the 6 shows that are spread across the two disks in the above scenario, you are virtually now doing 12 seeks (2 drives x 6 streams) for the 6 shows since EACH drive has to constantly seek the 6 different streams (recordings). In the situation where you have 2 separate drives, each disk is only required to constant thrash for the show that resides on the drive meaning that only in rare occasions will drive one have to constantly seek 6 streams/recordings and in most cases you will see a more balanced load of 3-4 streams/recordings per drive. This will REDUCE your thrashing of the hard drive which is really the issue you are most likely running into.

Seriously, I know my RAID's. RAID0 (which isn't a RAID at all) should NEVER be used in any environment as it will increase the failure rate exponentially without any real world advantages (yes synthetic benchmarks, but in real world experiences the slight decrease in load times/thoroughput is not enough to justify the high failure rates). The ONLY time it should even be considered is in large disposable files that need a large amount of thorough put, but as already stated even 6 HD streams only equals about 15-20 MB/s which is WELL below a modern hard drive of 70+ MB/s.

I realize you have made up your mind that RAID0 is the way for you, and good luck. Just be aware that you aren't going to accomplish what you think you will and in fact may actually experience even worse symptoms.

[mkanet]

I'm glad that I decided to go with RAID 0; despite how others theorized and analyzed about how RAID 0 shouldn't make a difference (in what I wanted to accomplish). I have clear proof that it does. I just _knew_ I was right; at least to some degree, about the effectiveness of RAID 0, but couldn't prove it. However, thank you guys for responding anyway.

I am able to record all 10 tuners at the same time and still comfortably FF/REW without any issues. I couldnt do close to that before on the same machine, same controller, same software. The only thing I changed was where I save to... a 2 disk RAID 0. In fact, even the disks I'm using for RAID 0 is practically identical to the 1 disk non-raid recording I used to do. Both non-RAID disk and RAID 0 disks were formated with 64k cluster size. I decided to go with 128K stripe size for the RAID 0; which worked out perfectly.

This thread might be useful to those interested in RAID 0 who aren't sure of the real-world benefts for sageTV. I can't imagine why someone else using RAID 0 under similar conditions would have different results.

[gplasky]

Congrats. I love it when a plan comes together.

Gerry

[mkanet]

Here it is in action right now with all 10 tuners... while I FF/REW Angela's Ashes movie.

Originally Posted by gplasky

Congrats. I love it when a plan comes together.

Gerry

[Fuzzy]

Glad it's working out for you. I did used to run 4 drives in RAID5, and had great performance. however, that was with rather old, slow drives. I have since upgraded to faster drives, and more HD recording, and decided to just keep things simple and go with seperate recording directories. I'm only looking at 5 recordings, so it's not near as significant.

If, in the future, Sage smartens up the drive selection algorithms, RAID0 really may be the best option. Second to that would be to set the encoders to alternate drives (merit 1 -> drive a, merit 2 -> drive b, etc.), but this may end up with drive a filling up too fast. Might require some scripts to level out drive usage during down times.

[mr_lore]

Thats a wild setup, great job!

As for RAID 0 being a benefit I would be hard pressed to come up with a reason why it wouldnt be beneficial, my setup below is taxed when recording two hd-pvr streams at once while playing back ripped 1080p content on just one other extender. I have yet to nail down exactly if this is disk performance (I think it is) or CPU/GPU but I have a feeling that my deal-of-the-day SATA 500GB primary recording drive is being maxed out as the 1080p conent becomes choppy dropping audio and video about every 2 minutes during this situation.

I've got a 2 x 60GB SSD RAID 0 in my primary desktop setup that I plan to migrate to my sage server soon when SSD technology settles down and 200MB sustained write/read becomes affordable in 100GB+ sizes. My only fear then would be wear and tear on the drives as they would be under near constant write.