SSD Are Me!

[sclitheroe]

Originally Posted by Rathji

Why would you want to do such a thing? Is your porn loading too quickly?

Because instead of your CPU sitting idle waiting for the data to come up from disk, you can put it to use to actually improve throughput rates, reduce latency on standard drives (not as big an issue on SSD’s), and even extend the lifespan of SSD drives.

If you have a fast CPU there isn’t much downside.

[photon]

Originally Posted by sclitheroe

In theory, you’ve also improved SSD lifespans, since fewer blocks need to be written to disk to store the same amount of data

Depends on the SSD I think, like I said the Sandforce controllers already do this, and are tuned in their wear leveling algorithms based on a assumed ratio of compressible data, compressing all the data first is going to throw this out of whack.

Enough to make a huge difference, who knows.. the OCZ support forums though the consensus seemed to be avoid compression for Sandforce drives.

[photon]

Originally Posted by FanIn80

Here's post-compression rating... (compare to the one I posted earlier in this thread, from when I first got the drive)

-----------------------------------------------------------------------
CrystalDiskMark 3.0 x64 (C) 2007-2010 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :   248.753 MB/s
          Sequential Write :   109.504 MB/s
         Random Read 512KB :   175.341 MB/s
        Random Write 512KB :   107.689 MB/s
    Random Read 4KB (QD=1) :    13.960 MB/s [  3408.2 IOPS]
   Random Write 4KB (QD=1) :    23.324 MB/s [  5694.4 IOPS]
   Random Read 4KB (QD=32) :   158.308 MB/s [ 38649.3 IOPS]
  Random Write 4KB (QD=32) :    99.722 MB/s [ 24346.3 IOPS]

  Test : 100 MB [C: 42.7% (63.6/148.9 GB)] (x5)
  Date : 2010/10/22 17:05:41
    OS : Windows 7  [6.1 Build 7600] (x64)

Looks like my performance did drop, although not to noticeable levels. I guess I have to weigh the performance loss with the 9GB storage gain. In reality, it's not like I even needed the extra hard drive space...

Hmmm...

-----------------------------------------------------------------------
CrystalDiskMark 3.0 x64 (C) 2007-2010 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read :   218.629 MB/s
          Sequential Write :    79.740 MB/s
         Random Read 512KB :   208.292 MB/s
        Random Write 512KB :    77.001 MB/s
    Random Read 4KB (QD=1) :    28.319 MB/s [  6913.7 IOPS]
   Random Write 4KB (QD=1) :    71.050 MB/s [ 17346.2 IOPS]
   Random Read 4KB (QD=32) :   163.669 MB/s [ 39958.2 IOPS]
  Random Write 4KB (QD=32) :    72.369 MB/s [ 17668.2 IOPS]

  Test : 50 MB [C: 67.4% (75.3/111.7 GB)] (x2)
  Date : 2010/10/22 17:56:14
    OS : Windows 7  [6.1 Build 7600] (x64)

Crystal Disk Mark doesn't give a picture across the drive though, HD Tune Pro will show performance across the whole drive, or HD Tach.



This is actually interesting, this was worse for me a few weeks ago, looks like my drive is slowly coming back in performance.. With the Sandforce drives the trim / garbage collection is very slow, but this is a lot better than it was. My whole drive was down at that 160MB/s level almost.

What's up with your Random QD=1 scores? Did you let Windows partition the drive, your partition alignment looks messed up.

[sclitheroe]

Originally Posted by photon

Depends on the SSD I think, like I said the Sandforce controllers already do this, and are tuned in their wear leveling algorithms based on a assumed ratio of compressible data, compressing all the data first is going to throw this out of whack.

Enough to make a huge difference, who knows.. the OCZ support forums though the consensus seemed to be avoid compression for Sandforce drives.

Oh crazy, I had no idea stuff like this was being implemented at the controller level. Interesting stuff...

[Rathji]

Originally Posted by sclitheroe

Because instead of your CPU sitting idle waiting for the data to come up from disk, you can put it to use to actually improve throughput rates, reduce latency on standard drives (not as big an issue on SSD’s), and even extend the lifespan of SSD drives.

If you have a fast CPU there isn’t much downside.

I can't believe I never realized that myself. For some reason I always associated compression with poorer performance. I think the last time I even considered it was when I was using a 750MB HDD.

[FanIn80]

Originally Posted by sclitheroe

Because instead of your CPU sitting idle waiting for the data to come up from disk, you can put it to use to actually improve throughput rates, reduce latency on standard drives (not as big an issue on SSD’s), and even extend the lifespan of SSD drives.

If you have a fast CPU there isn’t much downside.

Yeah, I read somewhere that compressing an SSD drive can actually increase performance. The theory was that you're moving a smaller file around, and that the decompress speed is actually faster than the transfer speed...

I didn't even think about the stuff you just mentioned.

Originally Posted by photon

What's up with your Random QD=1 scores? Did you let Windows partition the drive, your partition alignment looks messed up.

You know... I saw you talk about this in a different thread, and I meant to ask you what was up with that. I have no idea about partition alignment or how to do it (or even what the benefit is). I mean, I understand what partitions are and all that, but I have no clue about the alignment side of it.

I did let Windows partition my drive. Whenever I reinstall, I just select the old volume and delete it... and then tell Windows to create a new one and install to it.


PS. I love these SSD discussions. It seems like there's always something new to learn!

[FanIn80]

[sclitheroe]

Originally Posted by photon

Depends on the SSD I think, like I said the Sandforce controllers already do this, and are tuned in their wear leveling algorithms based on a assumed ratio of compressible data, compressing all the data first is going to throw this out of whack.

Enough to make a huge difference, who knows.. the OCZ support forums though the consensus seemed to be avoid compression for Sandforce drives.

This is one reason why I think SSD storage needs to get away from the SATA standard and integrate better with the OS. SATA defines a block storage model based on head, sectors, cylinders, etc. Now, you have devices hiding compression, de-duplication, sector sparing (which is OK), background TRIM, etc.

I don’t expect the OS to know what is optimal for file system performance on these devices, but it makes it hard even for an IT guy to know - now I have to inspect every chipset on every drive to know what a good recommendation for utilization is. (Or simply accept the default lowest common denominator - treat like a regular SATA)

There’s got to be a better way when it comes to SSD’s - they aren’t really devices that fit the SATA model anymore.

[FanIn80]

OK, so I did a bit of reading on partition alignment. It turns out that Windows 7 automatically aligns partitions on an SSD drive if you let it partition the drive for you.

My partition starting offset is 1,048,576 bytes.
1048576 / 4096 = 256

I'm still not 100% down with the benefits of doing this, but it appears I got lucky (or that Windows does a pretty good job of handling SSDs).

Edit: OK, now I get it. So if the partition's not aligned, the OS has to write a page twice (once for each half) which winds up doubling the number of writes you need. Well not really doubling, since it doesn't happen on every page, but enough that it makes a difference.

[MickMcGeough]

Newegg's got the 90GB Vertex 2 for $179.

http://www.newegg.ca/Product/Product...-20227601-L02C

Soooo tempted. Striped SSDs...

[MickMcGeough]

Just replaced my X-25M with a larger Sandforce SSD in my MBP and did some benchmarking. Interestingly, while the Sandforce drive has better write speeds, the X-25M still beat it in read speeds (which for most people is the more important of the two).

If anyone's interested in the X-25M it's up in the Buy/Sell section: http://forum.calgarypuck.com/showthread.php?t=98677

[sclitheroe]

Originally Posted by MickMcGeough

(which for most people is the more important of the two).

You sure about that? Everyone says read is more important, but in my experience with OS X, you'll write out almost as much daily as you read. For example, in the last 20 hours of uptime, my MBP has read 4 gigs, written 3.7 gigs. That's general office use, surfing, and a virtual machine running for VPN and remote desktop connections only.

You can see these numbers in Activity Monitor on the disk tab. I bet you'll find writes are closer to 50% of disk activity than you expected. I'd be interested if other people posted their numbers. I don't know if there's an easy way to get the number from a Windows box, but Mac users can easily check via Activity Monitor.

[MickMcGeough]

Originally Posted by sclitheroe

You sure about that? Everyone says read is more important, but in my experience with OS X, you'll write out almost as much daily as you read. For example, in the last 20 hours of uptime, my MBP has read 4 gigs, written 3.7 gigs. That's general office use, surfing, and a virtual machine running for VPN and remote desktop connections only.

You can see these numbers in Activity Monitor on the disk tab. I bet you'll find writes are closer to 50% of disk activity than you expected. I'd be interested if other people posted their numbers. I don't know if there's an easy way to get the number from a Windows box, but Mac users can easily check via Activity Monitor.

Reads: 1,821,121
Writes: 218,455

That can't be right... that's what it says though. I've got a dozen or so webapps running at any time with a SOLR instance, maybe that's really read-heavy?

I wasn't necessarily implying that most computers read drastically more than they write anyway. Just that for most people, the perceived performance increase from an SSD drive will be measured in bootup time and application startup time, and impacted more by fast reads.

[silentsim]

Reads in: 9007525
Writes Out: 5739515

[sclitheroe]

What kind of byte counts - that's the more interesting number when considering the read to write ratio, as read and write operations can vary widely in how many bytes they handle each.

[sclitheroe]

Originally Posted by MickMcGeough

Reads: 1,821,121
I wasn't necessarily implying that most computers read drastically more than they write anyway. Just that for most people, the perceived performance increase from an SSD drive will be measured in bootup time and application startup time, and impacted more by fast reads.

That's impacted more by the reduction in latency though - most of the boot and application load process involves lots of random i/o, which is what kills throughput on mechanical drives. Very few people actually need the kind of sustained sequential throughput that any drive is capable of, mechanical or solid state

[FanIn80]

Is there a guide somewhere for aligning the partition properly in OSX? I've Googled it, but there's so much info to sift through, all I'm really looking for is a way to do it quickly so I can get back up and running.

[photon]

I've never seen one unfortunately. Maybe check the OCZ forums, they're pretty active.

[DownhillGoat]

Does anyone have any experience with the Seagate Momentus XT? It's a hybrid drive. I want to get an SSD for my MBP, but don't want to lose the optical drive, and don't want to be limited to little storage space. Thinking this might be a good route to take but I was hoping others had some insight.

[photon]

The reviews that I read seem to be positive, not as fast as an SSD but does give a nice boost over regular drives.