It is currently Mon Sep 15, 2014 11:26 pm

All times are UTC - 8 hours




Post new topic Reply to topic  [ 15 posts ] 
Author Message
 Post subject: How important are random writes on SSDs
PostPosted: Wed Jan 05, 2011 7:45 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
HFat wrote:
dhanson865 wrote:
For now lets say the cut off is >30MB/s in random writes at low queue depth and small write sizes.

This is a bizarre metric and quite a high bar to set. I doubt all the drives you listed can sustain that in realistic conditions. We know there are conditions in which they fail to deliver.
More to the point, what purpose could such a capability serve? "low" and "small" are subjective so I'm not sure what you mean exactly but have you taken a look at how much random writes real consumer systems (not servers with many users) issue? By focusing on a non-issue, you're missing real performance bottlenecks.

I'd also like to see your source for the performance of the Samsungs.


I'll be back to reply to this shortly, just going to link from the other thread first.

Ok, "low" and "small" are subjective. The reason I said low and small is because Anandtech tests at QD = 1 but Tom's hardware tests at QD = 3 but the benchmarks are similar in other respects the the order of drive results is comparable. I'm sure there will be some site out there that will test at QD=2 or QD=4 or QD=5. I don't know how high the QD can be raised before it starts changing the results drastically but QD=32 is definitely not low and changes the game quite a bit.

small, oy vey, I just said small because I was lazy and didn't want to specify the size in B or KB. Please don't kill me for being inexact. The exact test specs I am looking at vary by site but here are some examples:

Iometer - 4KB Random Write, 8GB LBA Space, QD = 3 (reported in Average MB/s - Higher is Better)

CrystalDiskMark 3.0 x64 Random Write 4 KB (QD=1) [MB/s] Test : 4000 MB, Repetition: 5x (again I'd assume this was reported in Average MB/s and I'm sure Higher is Better)

There may be other tests that would fit into the small low queue depth window but even if it is only these two I thought it'd be easier for the average reader to not have to see me specify all the arcane test procedures. The pricing thread is a pricing thread. It's a general overview for people looking to make a quick comparison. Think of it as a message thread equivalent of the conclusion page of a SSD roundup.

I just thought it'd be better to not pen myself in and need to edit the first post of that thread every time a new review came out that changed the nature of a benchmark by one small parameter but still gave comparable results.

I think I'll end this post here and put answers to the rest of your concerns in two more posts.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Wed Jan 05, 2011 8:27 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
Samsung 470 reviews should probably be in a thread about specifically the Samsung 470 SSD but since I've started this thread primarily to discuss random write performance with you I am willing to clutter it slightly.

The first review of the Samsung 470 I read was at storagereview. See http://www.storagereview.com/samsung_47 ... 5pa256hmdr but I'm not a fan of their testing methods as they don't test sandforce drives with random data (ignoring a real world performance drop), their graphs are confusing (often the labels and bars aren't on the screen at the same time so I have to scroll a lot to see which drive is which in a roundup), and their numbers don't compare easily vs other sites. I do however like some aspects of their coverage and I tend to check storagereview before some of the other newer sites.

I read a bunch of other reviews that I'm not sure which was which at this point so I'll let you google "samsung 470 review" before I go on to one I remember specifically

Now here is a fun one

Roundup: The Best SSDs For Enthusiasts
http://www.tomshardware.com/reviews/sam ... ,2783.html

Ironic that they would use samsung-470 and sandforce in the roundup URL. Anyway here is a png from that roundup
Attachment:
crystaldiskmark_sequential_random_write_4kb_qd_1.png


and the equivalent from Anandtech
Attachment:
Iometer-vetexplus34081.png


Yes it is true that the performance between the two isn't identical even with the same drive controller but the relationship between slow drives and fast drives is sufficient.

Now I'm sure the Samsung 470 isn't perfect. The SSD magician is still in beta, the firmware is still in need of a few tweaks but it's early in this drives lifecycle. I expect it to do better when they update the firmware.

Oh and I'm sure you notice how all the Kingston drives drop to the bottom of the chart from Anandtech.

Full disclosure I don't work for any drive manufacturer or have any business relationship with any drive manufacturer/distributor/retailer/etcetera. I hope anyone that is willing to argue in public about the performance of a drive that doesn't do well on a specific test scenario would be willing to disclose their status as well. I only say this since this you've been attacking my paying attention to random writes across multiple threads.


You do not have the required permissions to view the files attached to this post.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Wed Jan 05, 2011 12:47 pm 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
OK so lets talk SSD performance in general a bit.

I'm like to assume you've read at least these articles

Quote:
Intel X25-M SSD: Intel Delivers One of the World's Fastest Drives
by Anand Lal Shimpi on 9/8/2008

The SSD Anthology: Understanding SSDs and New Drives from OCZ
by Anand Lal Shimpi on 3/18/2009

The SSD Update: Vertex Gets Faster, New Indilinx Drives and Intel/MacBook Problems Resolved by Anand Lal Shimpi on 3/30/2009

Intel's 34nm SSD Preview: Cheaper and Faster?
by Anand Lal Shimpi on 7/21/2009

Intel X25-M G2: Dissected and Performance Preview
by Anand Lal Shimpi on 7/22/2009

Intel Forces OCZ's Hand: Indilinx Drives To Drop in Price
by Anand Lal Shimpi on 7/23/2009

The SSD Relapse: Understanding and Choosing the Best SSD
by Anand Lal Shimpi on 8/30/2009

The SSD Improv: Intel & Indilinx get TRIM, Kingston Brings Intel Down to $115 by Anand Lal Shimpi on 11/17/2009

OCZ's Vertex 2 Pro Preview: The Fastest MLC SSD We've Ever Tested
by Anand Lal Shimpi on 12/31/2009

OCZ’s Vertex Limited Edition Review & SSD State of the Union
by Anand Lal Shimpi on 2/19/2010

Western Digital SiliconEdge Blue Review: WD Enters the Consumer SSD Market by Anand Lal Shimpi on 3/3/2010

Intel's X25-V & Kingston's 30GB SSDNow V Series: Battle of the $125 SSDs by Anand Lal Shimpi on 3/19/2010

6Gbps SATA Performance: AMD 890GX vs. Intel X58/P55
by Anand Lal Shimpi on 3/25/2010


and thus know what I mean when I use the phrase "the four corners of SSD performance".

Marketing types love to promote big numbers because bigger is better is more easily understood by the masses. So lets look at the "four corners" and see which play into the big numbers game. Keep in mind that in 2008 Anand didn't test the "four corners" at all. It was spring 2009 when Anand finally started testing SSDs in depth. I'll post some highlights for a time line

2009 spring Anthology
seq read: 250 MB/s OCZ Vertex
seq write: 195 MB/s OCZ Summit
random read: 55 MB/s Intel Gen 1
random write: 23 MB/s Intel Gen 1 (I'm ignoring the SLC X25-E)

and we get the quote "Random write performance is merely one corner of the performance world. A drive needs good sequential read, sequential write, random read and random write performance. The fatal mistake is that most vendors ignore random write performance and simply try to post the best sequential read/write speeds; doing so simply produces a drive that's undesirable. "

2009 Summer
seq read: ~260 MB/s Intel G1 and G2 160GB
seq write: 177 MB/s OCZ Summit*
random read: 58 MB/s Intel Gen 2
random write: 34 MB/s Intel Gen 2

* what changed to make the Summit slower? Probably a moot point but feel free to research it if you are curious.

anyway we get this quote "There are four basic pillars to SSD performance that I like to look at: random read, random write, sequential read and sequential write speed. A good SSD must be strong in all four categories, but some are more noticeable than others. Random read and write speed, particularly of small files (e.g. 4KB) are normally what make our desktop hard drives feel so slow. These random operations are everything from file and table updates to search queries and loading applications; they aren't random over the entire space of the disk but they are random enough to bring conventional hard drives to their knees."

2009 Fall SSD Relapse
seq read: ~260 MB/s Intel G1 and G2 160GB
seq write: 186 MB/s Patriot Torqx
random read: 64 MB/s Intel Gen 2 **
random write: 36 MB/s Intel Gen 2 (41 MB/s Intel Gen1)**

** The seq read scores haven't changed but the random got better.

2009 winter SSD Improv
seq read: ~260 MB/s Intel G1 and G2 160GB
seq write: 186 MB/s Patriot Torqx
random read: 64 MB/s Intel Gen 2 **
random write: 39 MB/s Intel Gen 2 (41 MB/s Intel Gen1)**

** The seq read scores haven't changed but the random got better again.

2010 spring 6Gbps SATA Performance (C300 used on all tests)
seq read: 343 MB/s Gigabyte X58 (Marvell controller) 6Gb/s
seq read: 268 MB/s Intel X58 (Intel controller) 3Gb/s

just here to show where the 3Gb/s limit kicks in in.

2010 spring SSD State of the Union
seq read: ~268 MB/s Crucial C300 (337 MB/s if you use 6Gb/s SATA)
seq write: 206 MB/s Crucial C300 (253 MB/s if you allow sandforce in the mix without randomizing data)
random read: 77 MB/s Crucial C300
random write: 39 MB/s Intel Gen 2 (51 MB/s if you allow sandforce in the mix without randomizing data)

Notice how if raw speed is all you want the C300 wins in the big numbers game. But on the task that slows down the drives the most Intel/Sandforce have controllers that handle the load better.

2010 Winter Kingston SSDNow V+100 Review
seq read: ~227 MB/s Kingston SSDNow V+100***
seq write: 193 MB/s Kingston SSDNow V+100***
random read: 79 MB/s Crucial C300
random write: 120 MB/s Sandforce 120GB (164 MB/s if you allow sandforce in the mix without randomizing data)

*** what, why is 227 MB/s in this review better than 268MB/s in other reviews? HMMMMM, maybe its because in the prior years reviews sustained reads and writes were tested with 2MB but in this review they switched to 128KB. Nothing like changing the rules of the game for a single review to make a specific product look better. Yep, there are 3 kinds of lies. Lies, Damn Lies, and Statistics.

I mean sure the 2MB read test was bottle necked by 3Gb/s SATA but is it any more valid to switch to 128KB? Maybe if I get the time I'll see if I can find comparable 2MB read and write numbers for the Kingston V+100 to see where it stands vs the older benchmarks.

So lets try a timeline by metric

Seq Read
250 MB/s OCZ Vertex 2009 spring Anthology
260 MB/s Intel G1 and G2 160GB 2009 Summer
260 MB/s Intel G1 and G2 160GB 2009 Fall SSD Relapse
260 MB/s Intel G1 and G2 160GB 2009 winter SSD Improv
268 MB/s Crucial C300 2010 spring SSD State of the Union
227 MB/s Kingston V+100 2010 Winter Kingston SSDNow V+100 Review

Obviously the Intel and Crucial drives didn't get that much slower in a few months time. The test was changed.

Seq Write
195 MB/s OCZ Summit 2009 spring Anthology
177 MB/s OCZ Summit 2009 Summer
186 MB/s Patriot Torqx 2009 Fall SSD Relapse
186 MB/s Patriot Torqx 2009 winter SSD Improv
206 MB/s Crucial C300 2010 spring SSD State of the Union
193 MB/s Kingston V+100 2010 Winter Kingston SSDNow V+100 Review

again the test changed for the V+100 review

Random Read
55 MB/s Intel Gen 1 2009 spring Anthology
58 MB/s Intel Gen 2 2009 Summer
64 MB/s Intel Gen 2 2009 Fall SSD Relapse
64 MB/s Intel Gen 2 2009 winter SSD Improv
77 MB/s Crucial C300 2010 spring SSD State of the Union
79 MB/s Crucial C300 2010 Winter Kingston SSDNow V+100 Review

the random tests didn't change for the new review

Random Write
23 MB/s Intel Gen 1 2009 spring Anthology
34 MB/s Intel Gen 2 2009 Summer
36 MB/s Intel Gen 2 2009 Fall SSD Relapse
39 MB/s Intel Gen 2 2009 winter SSD Improv
41 MB/s Intel Gen 2 2010 spring SSD State of the Union
120 MB/s Sandforce 120GB 2010 Winter Kingston SSDNow V+100 Review

in this test the Sanforce drive was finally tallied with random data so I allowed it into the list. But even if I didn't the Crucial C300 made 110 MB/s and the Intel Gen2 made 46 MB/s while the Kingston V+100 only made ~5 MB/s.

Man that firmware update sure did a job of improving C300 random write performance from 15 MB/s in the SSD State of the Union to 110 MB/s in the newest review. Obviously the early firmware wasn't ready for prime time.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Last edited by dhanson865 on Wed Jan 05, 2011 3:58 pm, edited 1 time in total.

Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Wed Jan 05, 2011 1:43 pm 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
Now lets put this in perspective vs a few hard drives and some lower capacity SSDs

Seq Read
100-120 MB/s Hard Drive (Seagate, Samsung, WD doesn't matter much which you compare vs)

185 MB/s Intel G2 40GB (Plenty fast enough to beat a traditional hard drive)
257 MB/s Intel G2 80GB

250 MB/s OCZ Vertex 2009 spring Anthology
260 MB/s Intel G1 and G2 160GB 2009 Summer
268 MB/s Crucial C300 2010 spring SSD State of the Union

Basically I never pay attention to Seq Read performance as it isn't an issue with any SSD worth buying. The 40GB Intel is the only Value drive that has even slightly slower read speeds. C300 still does seq reads at the limit even at the cheaper 64GB capacity.

Seq Write
42 MB/s Intel Gen 2 40GB
73 MB/s Sandforce 1200 controller 40GB drive
82 MB/s Intel Gen 2 80GB
100-110 MB/s Hard Drive (many drives are near this performance level with writes always being slower than reads)
102 MB/s Intel Gen 2 120GB/160GB
137 MB/s Sandforce 1200 controller 120GB drive

177 MB/s OCZ Summit 2009 Summer
186 MB/s Patriot Torqx 2009 Fall SSD Relapse
206 MB/s Crucial C300 2010 spring SSD State of the Union

Write speed is almost always slower on lower capacity drives. More so with some controllers than others but generally true. Not much you can do about this one so I tend to ignore it as well.

Random Read
0.3 to 0.7 MB/s on hard drives

20 MB/s Kingston V+100

45 MB/s Sandforce 40 GB
50 Mb/s Sandforce 120GB

55 MB/s Intel Gen 1 2009 spring Anthology
58 MB/s Intel Gen 2 2009 Summer
64 MB/s Intel Gen 2 2009 Fall SSD Relapse
77 MB/s Crucial C300 2010 spring SSD State of the Union
79 MB/s Crucial C300 2010 Winter Kingston SSDNow V+100 Review

Capacity isn't as important for this one but a good controller makes a big difference.

Random Write
2 MB/s or less on hard drives
5 MB/s or less on Kingston SSDnow drives

38 MB/s Intel Gen2 40GB
46 MB/s Intel Gen2 120GB/160GB

68 MB/s C300 64GB
71 MB/s Sandforce 40GB
110 MB/s C300 128GB
120 MB/s Sandforce 120GB

Again lower capacity SSDs have lower write speeds but now the controller can be a bottle neck as well.

OK now lets look at these again just weeding out worst case vs best case for the SSD but leaving a HD number in by comparison

Seq Read
100-120 MB/s Hard Drive (Seagate, Samsung, WD doesn't matter much which you compare vs)
185 MB/s Intel G2 40GB (Plenty fast enough to beat a traditional hard drive)
268 MB/s Crucial C300 2010 spring SSD State of the Union

Even the cheapo SSD is way faster than the hard drive and the spread between bottom SSD is 70% of the top SSD speed. Not much spread.

Seq Write
42 MB/s Intel Gen 2 40GB
100-110 MB/s Hard Drive (many drives are near this performance level with writes always being slower than reads)
206 MB/s Crucial C300 2010 spring SSD State of the Union

OK, slowest SSD is slower than a hard drive and the spread vs top ssd has the low one at 20%. But this is unavoidable with low capacity low cost SSDs. Pay for a higher capacity drive and the low write speeds go away. Even if you get the 40GB Intel G2 it's still faster than hard drives in real world use thanks to the improvements in the other three metrics.

Random Read
0.3 to 0.7 MB/s on hard drives
20 MB/s Kingston V+100
79 MB/s Crucial C300 2010 Winter Kingston SSDNow V+100 Review

The slow SSD here is 25% of the speed of the top SSD but this one is avoidable, you can buy a similarly priced SSD and beat the Kingston drive on this metric.


Random Write
2 MB/s or less on hard drives
5 MB/s or less on Kingston SSDnow drives
120 MB/s Sandforce 120GB

The slow SSD here is 4% of the fast SSD and you can easily beat this with even a cheaper SSD if you buy the right controller.


Hopefully seeing these recaps will help you understand the mindset of why I pay more attention to random read/write benchmarks and why I pay the most attention to random write. The spread on random write speed is the widest of the 4 fundamental performance metrics and thus the easiest one to use to separate the wheat from the chaff.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 9:22 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
So lets look at what can slow down your boot time, application loads, or affect general performance because it's causing random reads and random writes in the background on a Windows PC

* Windows Critical Updates
* Office Updates
* Antivirus updates
* Antivirus scans
* Adobe Flash Player Updates
* Adobe Reader (PDF) Updates
* Apple Quicktime/iTunes Updates
* Blackberry software
* Google Desktop
* Sun Java
* Weather widgets
* Stock Tickers
* Undetected Malware and Viruses
* Application preloaders for a Web Browser, Office Suite, Adobe Reader or other application that you would manually launch

and things that might cause continuing activity in the background that may or may not be loaded at startup (includes any from the list above that aren't loaded at startup on some PCs)

* Email Client
* Web browser (not uncommon for someone to have dozens of tabs open with pages refreshing in the background)
* BitTorrent Client or any other file sharing client
* Folding@Home
* Seti@Home, MilkyWay@Home
* GIMPS (Prime number), any number of PI calculators
* Distributed.net (Encryption and optimal Golumb Rulers)


Why do these matter? Well the truth is even though you and I may be smart enough to optimize our PCs and keep the crud off the average user in a company or the average family member you might be supporting doesn't. So if you are making a purchasing decision and want to know how that SSD will perform in the real world after you walk away you better be sure it isn't going to get swamped by random access demands of all the junk people get on their PCs.

And even if you don't support other users you may actually decide you want the convenience of having automatically updating applications on your PC.

Now you might say wait a second even the junkiest SSD is way better than a average hard drive. I'd say sure but the average SSD is is way better than the junkiest SSD so why pay for junk?

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 11:46 am 
Offline
Moderator

Joined: Thu Oct 06, 2005 4:36 am
Posts: 4601
Location: Monterey Bay, CA
Nice series of posts. :D

_________________
1080p Gaming build: i5-4670K, Mugen 4, MSI Z87-G45, MSI GTX 760 2GB Gaming, 8GB 1866 RAM, Samsung Evo 250GB, WD Red 2TB, Samsung DVD burner, Fractal Define R4, Antec True Quiet 140 (2 front + rear) case fans, Seasonic X-560. 35-40W idle, 45-55W video streaming, 170-200W WoW, 200-230W Rift, 318W stress test (Prime95 + Furmark)


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 12:13 pm 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
Thanks Steve.

BTW I just came back to show what is coming this summer with newer SSDs

128KB Sequential Write (QD=3) 262.MB/s
128KB Sequential Read (QD=3) 493.4MB/s
4KB Random Read (QD=3) 186.5MB/s
4KB Random Write (QD=3) 162.7MB/s

Now this is the highest end sandforce SF-2582 controller (6Gb/s) that isn't out for retail sale yet but will be used in the OCZ Vertex 3 Pro and they'll surely put something slightly slower in the Vertex 3 that will still blow away anything from 2010.

And that's not counting the C400 or Intel Gen 3 drives which are also not out yet (though the C400 will be out very very soon).

So yes, when I say I'm raising the bar from ~10MB/s random write to ~30MB/s random write I don't think that is extreme. Keep in mind drives that beat 10MB/s random write existed in 2008, drives that beat 30MB/s existed in 2009.

Extreme would be to only consider 6Gb/s capable controllers or only consider controllers released within the last 12 months. By the end of the 2011 calendar year the Intel G2 drives that will still be in my price list thread will be 26 months old (easily bought in fall 2009). I'm just cutting the slower drives at the turn of the calendar year instead of waiting until spring when the next generation of drives will be all the rage.

2011 Will redefine what I and most everyone else considers reasonable for SSD performance.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 12:45 pm 
Offline

Joined: Thu Jul 03, 2008 4:27 am
Posts: 1728
Location: Switzerland
I think it is best that I do not comment on the content of dhanson865's posts at this point. But I will briefly note that they do not explain why his cut off point (as opposed to 1MB/s, 10MB/s or any other value) would be relevant to the performance of a consumer drive let alone the defining metric for its performance in the field.

The context for this thread is that SPCR has reviewed a few SSDs and recommended a model which has terrible random write performance (an order of magnitude under dhanson865's cut off point) over the drive which defined the cut.
For those wanting a more technical perspective, Anandtech has a famous series of SSD reviews which are generally worth reading. They have a proprietary synthetic benchmark which exagerates the impact of random access performance but in which drives with relatively bad random write performance still perform reasonably well. One of the drives dhanson865 frowns upon in fact clearly outperforms every other drive in one of those benchmarks (I of course expect this to change with the release of new models).


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 1:14 pm 
Offline

Joined: Mon Jan 11, 2010 8:19 am
Posts: 400
Location: Boston, MA, USA
or get Linux and cache more!

Funny, I would have thought fancy wear leveling schemes would significantly help random writes too, since "random" is no longer "random" if you don't care too much about fragmentation (ie clean it up later).
Edit: The reason I bring up wear leveling is because it requires an abstraction of the address space with look-up tables, etc. It just occurred to me that this doesn't help if you are writing less than the unit write size. Isn't this usually <10kB though?

_________________
Currently running Mint 14 MATE on all machines.
Desktop/Gaming: E8400 / 650ti + Accelero S1 / Antec Solo
Myth Backend + Home Server: D945GSEJT / WD15EADS / Hauppage 950Q / Home made case
Myth Frontend + HTPC: Sempron 140 / Asus M4A785-M
Notebook:AMD A4-3300M


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Thu Jan 06, 2011 1:38 pm 
Offline

Joined: Thu Jul 03, 2008 4:27 am
Posts: 1728
Location: Switzerland
andymcca wrote:
I would have thought fancy wear leveling schemes would significantly help random writes too

They seem to, as expected. The more sophisticated drives and the ones relying heavily on the TRIM instruction for performance seem to have much better random write performance.
Drives with slow random write performance may have a high write amplification (low endurance).

andymcca wrote:
the unit write size. Isn't this usually <10kB though?

The write size yes. But you can only write one way (zeroes to ones - or ones to zeroes if you define it that way) before resetting and the standard "reset size" for flash is apparently very large.


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Fri Jan 28, 2011 10:11 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
andymcca wrote:
or get Linux and cache more!


Or get a dedicated RAID controller with more on board cache (not cheap)

Or buy a drive that has more cache on the PCB which is becoming more common

early JMicron based SSDs had 8 KB or 16 KB of cache
early Indilinx based SSDs had 16 MB or 32 MB of cache
later Indilinx based SSDs and some of the newer JMicron based SSDs have 64 MB of cache
the Kingston V+100 has 128 MB of cache as do many other drives including the new Performance 3 Series from Corsair.
C300/C400 use 256 MB cache

Sandforce and Intel do things with their controller that reduce the need for cache and get away with much less.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Fri Jan 28, 2011 10:19 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
HFat wrote:
The context for this thread is that SPCR has reviewed a few SSDs and recommended a model which has terrible random write performance (an order of magnitude under dhanson865's cut off point) over the drive which defined the cut.


The context for this thread is SSDs in general.

If you like you can read viewtopic.php?f=7&t=49956 and see how antiSSD some posters were. Bar81 flip flopped over and over on how great / how horrible his SSD was before I ever bought one.

I've used SSDs now and the market has matured some but there are still tons of XP machines out there and buying a SSD is not a no brainer. Caveat Emptor indeed.

I'm just trying to document some concepts and avoid a little bit of the negative vibe that is so common on technical threads. Make a positive comment. Tell me where you would draw the line. Be constructive not destructive and help your fellow board members make an informed purchase decision.

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Sun Mar 27, 2011 6:13 am 
Offline
SPCR News Editor

Joined: Thu Feb 10, 2005 11:20 am
Posts: 2175
Location: TN, USA
Many SSD Benchmark Reviews Contain Flaws
http://blog.zorinaq.com/?e=29

_________________
.
Please put a country in your profile if you haven't already.
This site is international but I'll assume you are in the US if you don't tell me otherwise.
RAID levels thread http://www.silentpcreview.com/forums/viewtopic.php?p=388987


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Tue Apr 05, 2011 6:43 pm 
Offline

Joined: Tue May 18, 2010 1:26 am
Posts: 19
Location: Earth
A big thank you to dhanson865 for an amazingly insightful thread.


Top
 Profile  
 
 Post subject: Re: How important are random writes on SSDs
PostPosted: Wed Apr 06, 2011 7:07 pm 
Offline
Friend of SPCR

Joined: Mon Dec 12, 2005 3:29 am
Posts: 371
Location: Melbourne, Australia
I concur.

Right now, the contrast between slow and fast is bordering on ridiculous.

Image

QD 32

Image

Vertex 3 random write Link: http://www.anandtech.com/show/4256/the- ... ew-120gb/6


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 15 posts ] 

All times are UTC - 8 hours


Who is online

Users browsing this forum: Yahoo [Bot] and 2 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group