Garbage Collection on External SSD (that's rarely mounted)

[sarangiman]

Hello community,

First post on Anandtech

Been looking for an answer to this but can't find anything.

I want to set up an external SSD (using a siig USB 3.0 enclosure that supports SATA 6Gb/s) to host my Synthogy Ivory piano samples. W/ a FW800 RAID 0 configuration using 7200rpm drives, I can only load 24 'voices' at a time. My initial test w/ a USB 3.0 SSD allowed me to load 700 voices at a time!

This is just an 80GB library, so I'll use a 120/128GB drive. Since it's a piano library, the drive will mostly experience reads. But I do record audio to it every now & then as well (voice, via microphone, so AIFF or WAV files).

Now, this drive sits on my digital piano, NOT at my regular computer desk.

In other words, the drive will only be connected/mounted when I go to play piano. I.e. not often.

What will happen to performance of the drive over time for: (1) Samsung-like drive that does idle garbage collection; vs. (2) Sandforce-based Intel 330 that does more active garbage collection.

Anand worries about Samsung drives' performance over time since garbage collection is only done when the drive is largely idle. Well, seeing as how my drive in this scenario would never be idle (except when it's unmounted from the computer... during which time it can't do garbage collection anyway, correct?), would a Sandforce based drive like the Intel 330 be a better choice for this scenario?

Basically, my question is: can SSDs that perform idle garbage collection do so when unmounted (but still powered, say from a USB 3.0 hub)? My guess is no.

And furthermore, in such a scenario where the drive is unmounted 99% of the time, and undergoing heavy sequential reads (150-200MB/s) when mounted, would a Sandforce-based SSD be a better choice due to its active garbage collection?

Many thanks in advance for any help!
Cheers.

 

[Cerb]

You're worrying too much.

If the drive is turned on, and GC needs to be done, it'll do it. If it is turned on and reading, any GC that needs doing will get done. Background GC isn't a situation where it only works when it is not being used at all (though some old drives did exactly that, and had problems because of it). It will be acting in the background during use, as well; it will just get out of the way when writes come in, to prioritize their performance over all else.

The problem cases for GC of many SSDs are far towards the opposite end of the storage spectrum from what you are describing as your use case (unrelenting random writes). Also, if your initial test was that much faster than the HDD, there's simply no way they will be sequential reads. The fastest consumer SSDs top out at only about 3-4x the sequential speed of current HDDs. A 30x speedup clearly indicates random reading.

Either of those drives aught to be just fine for your needs, though the Samsung is all-around a [slightly] better performer.

 

[sarangiman]

If the drive is turned on, and GC needs to be done, it'll do it.

But even if the drive is not mounted/connected to the computer at all? In other words, say it's only getting its power from a USB hub... and the USB hub is NOT connected to the computer at all.

Sorry you're right it's random reads not sequential reads. Random reads went up some 40-60x when I went from spinning drives to SSD.

Thanks much for your help.

 

[Cerb]

But even if the drive is not mounted/connected to the computer at all? In other words, say it's only getting its power from a USB hub... and the USB hub is NOT connected to the computer at all.

It should...but, even if does not, there won't be that much of it for it to do. You'll copy those libraries, which will be easy, as they'll be fairly big writes, for the most part. Those files will mostly just sit there, being read. Your audio recording is going to be almost nothing for them, being small sequential transfers (if you went all-out and did 192/24, that's only around 0.5MB/s/channel, IIRC).

I mean, look:
http://www.anandtech.com/show/4863/the-samsung-ssd-830-review/6

The torture test performed random write passes on the whole drive, which is all but impossible, in the real world, without running an SQL DBMS at full disk utilization all day, with a write-heavy workload. Then, he went, "well that's not slow enough," and proceeded to extend the test out to a longer duration than had been needed for previous drives. Even so, it still maintained a 50 MB/s write, and in excess of 350MB/s read. You will never even come close to giving the drive 0.1% of that level of write stress.

 

[sarangiman]

Thank you Cerb. All seem good points. You're right that the library will only be read, & the audio recordings will be small.

I'm new to the SSD thing so I'm a little paranoid. For example, I recently also purchased to ~500GB SSDs for on-the-go photo editing: Crucial M4 512GB & Intel 520 480GB. Even though I've filled my Crucial M4 512GB drive a number of times, erased, then rewrote files, it doesn't seem to have dropped in performance since day 1; meanwhile, my Intel 520 480GB drive which has ~70GB free has already lost 100MB/s write performance (sequential & random) since just a couple days ago.

Both drives are less than 1 week old.

I guess this agrees with what you said -- that the Intel drives seem to have lower relative performance. But I wanted the Intel 520 for its reliability & 5 year warranty, especially since for that particular drive I'm not too worried about write speeds as read speeds are of paramount importance when reading tons of RAW files in Lightroom for editing. They do become important for when RAW files are flattened to TIFFs for work in Photoshop, where there'll be layers & layers & file sizes can quickly become in excess of 1GB. But in that case, I was comforted by the low write amplification of the Intel drives... and since I want this drive to last a long time, for many wedding shoots...

But maybe all in all I'm just overthinking all this

Just placed the order for the Samsung 830 128GB!

Thanks again for your help!

 

[Cerb]

meanwhile, my Intel 520 480GB drive which has ~70GB free has already lost 100MB/s write performance (sequential & random) since just a couple days ago.

That's one of the reasons you won't find us recommending even Intel's Sandforce (SF) drives over non-SF drives, though Intel did do much more than just rebadge them. SF assumed that cheap low-quality flash would be too good to pass up, and their controllers have features that let them write less to the flash itself than others, typically actually writing less actual data than the size of data coming in, when the data is easily compressible. But, whether it can maintain the like-new performance over time varies much more than with drives not using SF's controllers.

 

[sarangiman]

Is there a good reference I could read on why write performance would decline so much so quickly on SF drives?

Thinking now about the Samsung 830 512gb but the higher power consumption worries me in external enclosures.

Perhaps the Samsung 840 Pro is a better bet, as it appears to have lower power consumption?

Or again am I worrying too much (mostly interested in USB 3.0 enclosures for now, as I can use the really slim Seagate GoFlex Slim case & it's quite small & slick... also TB enclosures heat up like crazy, not so for the USB 3.0 enclosures I've tested).

Thanks again!

 

[sarangiman]

Actually, on that note, I haven't confirmed that the Seagate GoFlex Slim enclosure/adapter is actually SATA 6Gb/s (to USB 3.0), and I can't find any info about this anywhere on the internet.

However, in my XBench & Drive Genius 3 benchmarks, the read/write speeds of both the M4 512GB & the Intel 520 480GB drives are exactly the same for the Seagate Slim adapter (http://www.seagate.com/external-hard-drives/portable-hard-drives/performance/goflex-slim/) as for the SIIG SATA 6Gb/s to USB 3.0 enclosure (http://bit.ly/PvarJQ).

But it's always possible these drives aren't pushing the 3Gb/s limit? highest read speed measured is ~286MB/s, though over Thunderbolt that goes up to ~330MB/s for the 520 (Seagate Thunderbolt adapter). So I was assuming USB 3.0 was the bottleneck, not SATA II vs III.

Anyway, just ordered a specifically-labeled SATA 3Gb/s to USB 3.0 enclosure to see what sort of speeds it has next to the SIIG 6Gb/s to USB 3.0 & the Seagate Slim.

Given the general lack of Thunderbolt enclosures, I probably feel safer sticking with USB 3.0 anyway. For whatever reason though, 4K random writes always suffer considerably with USB 3.0 compared to Thunderbolt (4-5x drop); and of course sequential read/writes are about 100MB/s faster over Thunderbolt, but the latter doesn't matter too much to me.

 

[Hellhammer]

You have nothing to worry about.

Your library is only 80GB, which means nearly 40% of the drive will be empty. Why would performance degrade, there is plenty of free blocks to write to. On top of that, your data is mostly sequential, which isn't as harsh for the drive as random data is.

 

[Cerb]

Given the general lack of Thunderbolt enclosures, I probably feel safer sticking with USB 3.0 anyway. For whatever reason though, 4K random writes always suffer considerably with USB 3.0 compared to Thunderbolt (4-5x drop); and of course sequential read/writes are about 100MB/s faster over Thunderbolt, but the latter doesn't matter too much to me.

The enclosures add significant layers of abstraction (Thunderbolt far less than USB), which reduces performance compared to an internal drive. But, you said you already tested with a USB 3.0 SSD, which should have similar performance to putting a SATA one in an enclosure (most speedy factory ones are basically the same thing, electrically--they just save on manufacturing costs by using a single case and circuit board).

But, how often are you going to be doing 4K random writes? That's a pathological test case, to find out what happens when you go well outside of typical use cases--it's telling of performance differences, but it's not a common occurrence that you will write a bunch of random sectors, without pauses, larger writes, and many reads, in between. You are going to giving it 50+K/ea. sequential writes, on rare occasion, by the way I read it.

 

[sarangiman]

To be clear: I haven't tested any USB 3.0 external SSDs as packaged together by manufacturer (don't even know if these exist).

I ripped apart a Seagate GoFlex Slim USB 3.0 case & put a SSD drive in there. Also bought a SIIG SATA 6Gb/s to USB 3.0 enclosure & put SSD drives in that. I compared the same SSD drive(s) in both these enclosures and got similar results, leading me to believe that Seagate's SATA-->USB 3.0 adapter is compatible with SATA III, with the caveat that that conclusion is contingent upon the SSD drives I put in there being capable of benefiting from SATA III vs. II. Which is why I ordered a SATA II enclosure to see if there's a drop in speed compared to the aforementioned SATA-->USB 3.0 enclosures.

Good point on random 4K writes. That plus I think I'm willing to sacrifice 375MB/s for 275MB/s for the convenience (& slimness!) of the USB 3.0 enclosure.

But now seriously thinking about the Samsung 840 Pro given the recent drop of 100MB/s write performance w/ my Intel 520 drive.

I just can't find much info out there about people putting big SSDs (>400GB) in small TB or USB 3.0 enclosures. Some forums exist on macrumors talking about using SSDs in the Buffalo MiniStation, for example, but talk about reliability issues. I'm assuming this is due to power.

I myself seem to see minor issues -- like when I do large directory copies, every now & then they fail within the first 10 seconds, but so far every time on my 2nd try the copy goes over just fine. These are like 400GB transfers, typically running at 200MB/s, with periodic dips when I look at the performance using Activity Monitor.

Or every now and then the benchmark suites fail saying they can't read the test file. Again, they always work on the 2nd try.

Don't know if that's something I should worry about. Have any of you ever experienced this?

Thank you again.

 

[Cerb]

But now seriously thinking about the Samsung 840 Pro given the recent drop of 100MB/s write performance w/ my Intel 520 drive.

Totally different controllers. The Samsung is far less likely to have that kind of drop.

I just can't find much info out there about people putting big SSDs (>400GB) in small TB or USB 3.0 enclosures. Some forums exist on macrumors talking about using SSDs in the Buffalo MiniStation, for example, but talk about reliability issues. I'm assuming this is due to power.

Aside from Mac users, most people don't have TB ports, and hardly even know of it. USB 3.0 is still fairly new, and >400GB SSDs are expensive, and thus rare purchases. For example, I looked to see wha chip OWC was using, and any performance reviews of their enclosures, since they tend to target professional non-techies that need things that work, and I came up blank: only slow HDDs (no way to gauge bridge performance), and I could not find a picture of the underside of the PCB.

You get to be an early adopter!

I myself seem to see minor issues -- like when I do large directory copies, every now & then they fail within the first 10 seconds, but so far every time on my 2nd try the copy goes over just fine. These are like 400GB transfers, typically running at 200MB/s, with periodic dips when I look at the performance using Activity Monitor.

Or every now and then the benchmark suites fail saying they can't read the test file. Again, they always work on the 2nd try.

Don't know if that's something I should worry about. Have any of you ever experienced this?

No, but that seems an awfully lot like a bridge problem. With large transfers, it could be overheating. Most current USB 3.0 enclosures (using an ASMedia chip) seem to be limited to about 200MB/s, as well (the performance drops may or may not be hardware, if you are using OS X).

 

[sarangiman]

For example, I looked to see wha chip OWC was using, and any performance reviews of their enclosures, since they tend to target professional non-techies that need things that work, and I came up blank: only slow HDDs (no way to gauge bridge performance)

Yeah what's up with all these manufacturers making USB 3.0 & Thunderbolt mini drives & putting spinny drives in them? Not only that, 5400rpm spinny drives?!

I end up buying the drives just to use the enclosures then chucking the actual HDD.

No, but that seems an awfully lot like a bridge problem. With large transfers, it could be overheating. Most current USB 3.0 enclosures (using an ASMedia chip) seem to be limited to about 200MB/s, as well

Yes but the errors always occur within the first 5-10 seconds of starting the file transfer or the benchmark. The 2nd time, they always work (n of like 5 attempts?), & never fail near the end. If it were an overheating problem, it wouldn't occur at the beginning, right?

Also, wonder if the Seagate uses the ASMedia chipset? The actual SATA-->USB 3.0 board is tiny -- just the backend that plugs into the drive. Makes me wonder if its their own in-house thing.