VPCZ-1x SSD SPEED REDUX

This is something of a small clone of an old thread here on Z RAID0 SSD speeds and degradation over time which I cannot find after searching 10 pages back!

I bought a refurb Z13 from Sony several months ago, and the SSD (2X64 RAID0) seems to perform very well, but for one anomaly in the CrystalDiskMark results (shown below). Everything looks good to me except the 512K Random Read and write, which score at 120.8 and 58.7 respectively, both of which seem low compared to my other similarly configured Z (see below), but the 512K write speed is the one that's way off the mark.

These 512K read/write rates are fairly consistent when I re-run them every few weeks, though below the first set of results is a second run I did recently in which the random write went from 58.7 here to 132.0 while the random read stayed around 120.. But the lower write score is more representative of what I usually get on this SSD array. In fact, over the 5 trials per test it can go as low as 9.0 mb/s and usually recovers to something in the 50s or 60s.

Adding to the intrigue, on a virtually identical machine (but a Z11) I get very similar numbers as the ones shown below across the board, except those nutty 512K read/writes, which score 207.1 read, 179.2 write, eclipsing all of the runs done on the Z13. I have most of the common speed tweaks in place in both machines.

I guess my questions are obvious, but I will simplify my biggest concerns:

1. Why would the 512K Write, not so much the Read, score so consistently slow on the z13 I have while all other measures seem pretty good for the Z13's 64+64 array? And, what would you make about intra-test 512k write scores as low as 9 and as high as 132?

2. Why would my Z11, same array, score almost double on the 512K read and as much as 3X on the 512 write, while every other test came in about the same between the Z11 and Z13?

3. Does the change from Samsung to Toshiba memory account for this much difference?

4. What "real world" task does the 512K test simulate? Could the write score on this array be degraded due to anomalously high use of tasks that simulate a 512K write?

5. The Z13 is under wtty. Should I send it to SONY (no, no, please say NO!!!) and get it replaced? Would they replace it based on a CrysalDiskMark?

6. Any other test to suggest to cross-verify?

I should mention that the Z13 feels as snappy as the Z11 despite being way faster in that benchmark 512k read/write. I only test to chart any changes over time and because I will be selling one of these.

Thanks so much. I know there are a lot of questions there. Also, if someone can get me a link to the earlier thread and perhaps to a part where this kind of issue was discussed, that would be great for me and would lessen your time.

.


TWO CRYSTALDISKMARK RUNS FOR THE SAME SSDS ON A SONY Z13
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 302.270 MB/s
Sequential Write : 336.730 MB/s
Random Read 512KB : 120.883 MB/s
Random Write 512KB : 58.693 MB/s
Random Read 4KB (QD=1) : 12.944 MB/s [ 3160.2 IOPS]
Random Write 4KB (QD=1) : 32.333 MB/s [ 7893.7 IOPS]
Random Read 4KB (QD=32) : 31.432 MB/s [ 7673.8 IOPS]
Random Write 4KB (QD=32) : 67.038 MB/s [ 16366.8 IOPS]

Test : 1000 MB [C: 38.8% (40.5/104.4 GB)] (x5)
Date : 2011/06/23 9:45:55
OS : Windows 7 SP1 [6.1 Build 7601] (x64)


Sequential Read : 297.089 MB/s
Sequential Write : 322.242 MB/s
Random Read 512KB : 118.661 MB/s
Random Write 512KB : 132.863 MB/s
Random Read 4KB (QD=1) : 13.426 MB/s [ 3277.7 IOPS]
Random Write 4KB (QD=1) : 39.468 MB/s [ 9635.7 IOPS]
Random Read 4KB (QD=32) : 30.767 MB/s [ 7511.5 IOPS]
Random Write 4KB (QD=32) : 69.763 MB/s [ 17031.9 IOPS]

Test : 1000 MB [C: 34.4% (35.9/104.4 GB)] (x5)
Date : 2011/06/19 2:06:06
OS : Windows 7 [6.1 Build 7600] (x64)
------------------------------------------------------------------------------------------------
-----------------------------------------------------------------------
CrystalDiskMark 3.0.1 x64 (C) 2007-2010 hiyohiyo
Crystal Dew World : Crystal Dew World
-----------------------------------------------------------------------

 

Crystaldiskmark is everything but accurate IMO. Try ATTO and post your results here. Here are mine beside my Intel RST configs (sorry it's German).

 

Thanks, I'll run it and edit this post with results. Have a question only a noob should ask, but, I don't know how to get a graphic into a post I used to use Windows Snipping tool to cut it out, copy to clipboard and paste it in, but that isn't working for me - wah! Help, pls.

 

prtscr copies the entire screen into clipboard. Then you can paste into paint.

ctrl-alt-prtscr copies just the active window into clipboard. Then you can paste into paint.

 

Alt-PrintScr is enough for the active window, then upload the paint image somewhere (e.g. imageshack) and link it in your post.

 

Thanks much for all the coaching. Can't drop in my result, even though I uploaded it to Paint. Can copy the image and paste into the input box for this post, but when I submit, it just ignores my graphic. Help again! And to think I've been using computers continuously since 1973 ).

Anyway, I can explain in words the result I got which I think is pretty much in line with expectations for the size and config of my RAID array.

Two primary observations/questions:

1) My write speeds are reasonably comparable to Pyro's but my read speeds are only about 50% of yours. Is this explained in whole by the fact that you have 4X128GB and I have 2X64?

2) Once again I have one anomalous result, but this time it's a read speed at 16.0 K which dropped down to 124mb/s, then shoots back over 300, as it was at 8K, so no correlation. Is this one strange result just an a hiccup in the test?

So, do my results seem ok for a 6 month old Z13 with 128GB Raid0 SSD?

Thanks again for all

 

---bad post, replacement coming

 

Judging your explaination, it seems to be fine. Note that while benchmarking, most users won't run in a perfectly idle environment, meaning that each benchmark result may get influenced by background services and stuff that are accessing the drive while you are benchmarking. This results in hiccups and sometimes in questionable results. Sometimes, users are getting utopic numbers like 700MB/s in Crystaldiskmark which on the other hand is windows caching related and also not realistic for a Z.

In order to get a truly reasonable result (isn't this what we really want to know here?), one should run a row of benchmarks and average the results. IIRC, a dual raid setup like yours should come around 350MB/s read and 280-300MB/s write whereas a quad should get around 500MB/s read and 350MB/s write. All numbers are sequential (the high transfer sizes, 4-8MB) and samsung/toshiba drives should perform roughly the same.

Last weekend I routinely checked my RAID drive state and data integrity on my 1 year old Z (knocking on wood, no issues so far!) and gave my SSDs garbage collection some work to do. I filled my 512 gigs with giant text files and watched the drive overwriting old stuff with a drive sector viewer. Copying files first started at ~500MB/s and went down to constant 80MB/s after some gigabytes. I thought the garbage collection was inactive and something was wrong. I deleted all those dummy files, let my Z idle through the night and the screenshots above show the results the next day. Although, even today, if I transfer big files (3GB and up) of data, speed somewhat goes down. A 500MB file gets copied in literally one second.

Need to figure out with other Z owners more, perhaps I'll open another thread for that.

 

Thanks very much. Pardon my lameness, but I want to be sure I understand something. That process of overwriting files to fill the SSDs, do you do that to "exercise" the garbabge collection function of the SSDs or to test how they are doing? My confusion is that you said you stopped it and deleted all of the dummy files, then ran ATTO and found it was writing at 300 mb/s, not 80 as the viewer was showing. Was the SSD's built in garbage collection routine processing all night and that's why you then gained consistent 300mb/s write speeds the next morning?

Put another way, would you mind explaining what happened and why? I would have PM'd this, but I don't think I'm the only one who would benefit from understanding this better, especially if you are describing a routine that all should do periodically for the health of the non-TRIM's RAID0 SSDs.

 

I did not intend to exercise the GC, it was just a test. After I reinstalled windows from scratch and had some issues with installing RST (my firewall caused some weird errors) I had my SSDs running without Intel RST drivers. I then noticed that copying large files slows SSD speed down noticeably while smaller files do not, I thought there was something wrong with GC or GC did not engage due to the fact that RST drivers were not present. This was not the case and in benchmarks, my SSDs performed as fast as in the screenshots before and after I filled the drives. I did this to wipe out old data which might not be cleaned by the GC and deleted all dummies in order to force GC to engage during the night. Unfortunately, big files still cause excessive slowdowns (<90MB/s sequential) and I don't know the reason.

 

Can anyone compare speed for factory SSD 4x64GB after 18 month

I hope no bad

 

I believe these are excellent results! Just read posts by pyr0, below. He has quad-SSD array (4X128) in a Z11 and is the resident expert on most things Z related, and he stated that 500 mb read/350 write is peak performance for quad R0 array. Congratulations that your SSDs seem to have no deterioration in what must be at least 18 months of use!