Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Importance of random write speed?
- Thread starter Loa
- Start date
- Sort by reaction score
As important?
Nope. But still important. Copying a buttload of small to medium files is Random Write-ish. Copying two to one hundred 700MB movies is much more on the Sequential Write-ish side. Recording video is usually sequentially written but usually the data stream is much lower than the throughput of even a single drive. Same for audio. With audio this changes with the number of tracks being recorded simultaneously - and the format/method the recording software employs. Saving large application-data is sequential-ish. Saving or generating 2 to 100 files small from an app is Random Write-ish. Encoding a DVD in HandBrake is Sequential but again the data-stream is far less than the throughput of the drive.
When the data-stream is less than the throughput the storage unit and interface is capable of as in some of the cases named above, then "rotational latency" is important to consider in understanding what affects performance.
Hello,
Thanks for the replies guys.
I'll put the question in a more specific context: I'm really thinking of getting either the "new" X25-M 80GB or the Vertex 120GB SSDs for my OS drive. All of my data will be on my 2TB RAID0.
The only major difference between those 2 SSDs is the random write speed.
In the context of a OS drive, how would that affect speed/snappyness?
Thanks
Loa
Thanks for the replies guys.
I'll put the question in a more specific context: I'm really thinking of getting either the "new" X25-M 80GB or the Vertex 120GB SSDs for my OS drive. All of my data will be on my 2TB RAID0.
The only major difference between those 2 SSDs is the random write speed.
In the context of a OS drive, how would that affect speed/snappyness?
Thanks
Loa
From what I've read, 80% of a single user workstation storage activity is reading and most of that is small random reads.
The bulk of the rest is small random writes.
I've also read that it's not uncommon for there to be a queue depth of 80 on some operations (loading an app) so access time is a significant factor and this is one area where Intel is strong.
Agreed.
Random writes are still drastically higher than a HDD, but a slight difference won't matter.
Well it's not a small difference (Anandtech tests):
Vertex: 2.4MB/s
Intel's X25-M (the new one): 34.5MB/s
Loa
No, you're thinking of random reads. For random writes most 1TB drives are usually much faster than an SSD. For example my WD 1TB green drive is about 120 MB/s average (12 MB/s - 230 MB/s). That's almost three times faster than the x25-M.
Ouch! That's a pretty massive difference. Get the faster one.
If we all recall, it was the random writes that caused the first generation SSD "stutter" problems. I'll never forgive JMicron...
OT but I lay blame at the feet of the drive manufacturers who launched products that didn't work (using a controller that wasn't up to the task).
Back on topic...
I've seen some recommend the Intel for OS/Apps drive and the Vertex for file storage... however, if you are going to buy two drives then I think you'd be better off to buy two Intel's and run RAID0 for everything... effectively doubling the speeds.
If you have budget for only one drive, I think it's almost universally suggested that Intel get your money.
Could someone please elaborate on what exactly is random write? What kind of tasks specifically are you doing when random writing is involved.
Is it just a matter of file size?
Thanks.
Is it just a matter of file size?
Thanks.
Here goes an attempt at an explanation...
A random write is a small write to the drive to a seemingly random location. Updating a file system journal entry, writing out a page to disk, saving a snap shot of a file, caching an image from your browser, etc. the specific storage locations are all over the drive.... hence the term "random". Most OS related disk activity is of the random variety and goes on without much user interaction.
Traditional magnetic drives suck at random reads/writes because the head has to move across the drive surface for each write, thus introducing latency. SSD's are much better since there are no moving parts the system needs to wait for. However, the NAND flash used in SSD's also incurs a write block penalty which introduces latency to the overall operation, albeit much less than seek times on a magnetic drive.
A sequential write is the opposite... a large copy/save to the drive where several blocks of data are written in succession, one after the other, in adjacent locations on the disk. Traditional hard drives can be very fast at this operation because the head doesn't need to move much.
You should probably read this whole article, but this page in particular...
http://www.anandtech.com/storage/showdoc.aspx?i=3531&p=25
Random writes is a benchmark test write pattern. Usually it's not actually "random" but rather scattered in a butterfly pattern across the platter surface. They should probably call it scatter-writing and not random.
OS X (HFS) never ever does anything like what these tests do. You might be able to get it to happen if you created a million 1MB files to completely fill the HDD, then deleted every 500th file, and then tried to write a 1GB file onto that HDD. In OS X almost every write is a sequential write. OS X tries to always select a continuous space to write to. Scattered Reads (random reads) are a different story tho as it can't really choose where to read it from.
It is where it is and if you're loading files from scattered locations it will need to seek around.The Random Write test is designed to test write performance under conditions where rotational and seek latencies are massively abundant. Seek latency is the time it takes for the the R/W head to find the location it needs to move to and then to move there. Rotational latency is the time it takes after it's moved there for the controller and head to sync up with the data/logical-structure rotating underneath the head just prior to beginning the actual write or read operation.
I would say there are mainly two scenarios where random writes occur:
1. Copying many small files to your SSD (as said above) However, due to the speed of even "slow" SSD's the random read speed of the source drive is likely to be also a limiting factor.
Dramatic speed drops can occur if you use a SSD without cache, but I think the two drives you chose are well equipped.
2. Random-ish writes can occur if you use the SSD as cache/swap volume.
While having enough RAM installed in your computer almost completely avoids such operations by OS X, programs like Photoshop use a dedicated cache file which is almost always accessed when processing relatively large files.
In this case, also the sequential write speed of your drive can have an impact on performace.
I do not think that the random write speed difference of the two drives affects the system's snappyness in a noticeable way, remember that not only the drives but also OS X has a write cache, such that random writes below a certain size are written to the RAM or the SSD cache before they are actually stored in the flash memory, which speeds things up.
Conventional HDs have also write latencies. Many of these mechanisms developed for the reduction of the latency impact work also for SSDs...
1. Copying many small files to your SSD (as said above) However, due to the speed of even "slow" SSD's the random read speed of the source drive is likely to be also a limiting factor.
Dramatic speed drops can occur if you use a SSD without cache, but I think the two drives you chose are well equipped.
2. Random-ish writes can occur if you use the SSD as cache/swap volume.
While having enough RAM installed in your computer almost completely avoids such operations by OS X, programs like Photoshop use a dedicated cache file which is almost always accessed when processing relatively large files.
In this case, also the sequential write speed of your drive can have an impact on performace.
I do not think that the random write speed difference of the two drives affects the system's snappyness in a noticeable way, remember that not only the drives but also OS X has a write cache, such that random writes below a certain size are written to the RAM or the SSD cache before they are actually stored in the flash memory, which speeds things up.
Conventional HDs have also write latencies. Many of these mechanisms developed for the reduction of the latency impact work also for SSDs...
Those benchmarks are old. The Vertex is getting better garbage collection soon anyways through custom firmware.
http://www.pcper.com/article.php?aid=760
Random writes is a benchmark test write pattern. Usually it's not actually "random" but rather scattered in a butterfly pattern across the platter surface. They should probably call it scatter-writing and not random.
OS X (HFS) never ever does anything like what these tests do. You might be able to get it to happen if you created a million 1MB files to completely fill the HDD, then deleted every 500th file, and then tried to write a 1GB file onto that HDD. In OS X almost every write is a sequential write. OS X tries to always select a continuous space to write to. Scattered Reads (random reads) are a different story tho as it can't really choose where to read it from.
The Random Write test is designed to test write performance under conditions where rotational and seek latencies are massively abundant. Seek latency is the time it takes for the the R/W head to find the location it needs to move to and then to move there. Rotational latency is the time it takes after it's moved there for the controller and head to sync up with the data/logical-structure rotating underneath the head just prior to beginning the actual write or read operation.
So, if this is true would it be safe to say for normal operations the ocz drives are just as good as the intel's? What about in RAID 0? By normal I mean just as boot drives or downloading files and converting them, then moving to an external.
So the benchmarks go from 2MB/S and 35MB/S to 200MB/S and 250MB/S and you don't say to yourself: maybe they're testing something different here?
When most ppl think of random read/writes they mean 4K blocks of data. The chart from your article "start" at 64K and goes up.
Loa
I guess, I dunno. You would need to look at benchmarks to know for sure I think.
I guess when using an SSD for booting they're all good to about the same degree. I mean if one does 500 MB/s and the other does 750 MB/s that's only a 25% increase. Yet the 500 MB/s one is a 700% increase over a single 1TB drive which does far less than 100 MB/s. Typically to match an SSD for booting (small data-size reads) you need at least a 3-drive RAID0 of 1TB drives or better. For writing, SSDs are typically slower (or close to the same) than rotational drives of 1TB of larger.
What about if you were strictly encoding video then transferring it to a external drive? Or downloading data, converting it, then trasferring.
Encoding is a write operation and HDDs are faster for that than SSDs. But while drive speed does indeed impact encoding time (quite a lot actually) I think CPU speed may be even more critical.
Downloading is a similar deal. HDDs are faster at writing and a faster drive does indeed impact DL times but your line speed, ping, and etc., are much more critical.
Conversions typically allow the user to define the source and destination drives separately. So the best case would be reading from an SSD or SSD RAID0 and writing to a 1TB or greater drive or a RAID0 of the same.
For editing video I imagine it depends on the editor (FCP, Premiere, AFX, etc.) but I believe most of them are read intensive so SSDs would be better - and especially if the save or output drive was a rotational drive or RAID0 of the same.
I just ordered the Corsair P256, which is very well reviewed on Bare Feats. SSDs seem to shine as boot drives where you are mostly loading the OS, loading applications, etc. I'm hoping replacing my boot drive with the Corsair will make a noticeable difference in boot-up and launch times. Beyond that, I will be using my RAID 5 array for all file-related storage.