2x X25-V vs. everything else review.....

[Biker]

... over @ Anandtech

[One_Hertz]

So much for any faith I had in the anand storage bench.

[Computurd]

why would you say that?
is your drive not at the top?

[One_Hertz]

why would you say that?
is your drive not at the top?

Because 2x X25-V in R0 can not be near 2x faster than a single X25-M. They should be faster, but not 2x, which is absurd.

As soon as I see a large inconsistency like that, the whole benchmark loses validity in my mind.

Edit: to expand, the numbers don't make any sense. Specifically on the heavy trace. A lot of what they do is dependent on sequential writes, which is why we see drives high in that attribute near the top. The X25-M does poorly in comparison to other MLCs that do offer high seq writes, which makes sense. However, the fact that their 80MB/s write speed raid is suddenly 2x faster than a single X25-M which is also 80MB/s makes zero sense whatsoever. The extra read speed of the raid should help, but nowhere near 2x. They either messed something up in this particular benchmark or the whole thing is crap.

Edit2: to expand further, comparing both setups:

Seq read - 370 vs 255
Seq write - 80 vs 80
4kb r write - 60 vs 40
4kb r read - 60 vs 60

The controllers they are using (X25-V and X25-M) are very similar so the speed of reading + writing at once (also very important for the heavy trace, in fact this is why X25-E is the top drive) should be similar.

With that in mind, it does not follow that their trace should spit out 2x higher numbers in favor of the RAID. I would think around 25% higher would be appropriate.

[UObean]

Any test that mainly utilizes sequential reads, the two R0 drives would be twice as fast as an X-25m. Sequential reads are pretty linear in their scaling up to a certain point.

[Computurd]

hmm strange. steve-ro thread here seems to indicate very close results to what they posted up on anandtech.

[alfaunits]

What test is the Intel X25 R0 2x faster than a single? Except sequential?

The aligned writes vs. regular writes seem so good on the new drives. How TF do they even top the reads on the same drive is strange. Could reads also need aligning...

[NeedMoMegaHurtZ]

But there is something funky going on with the heavy workload trace...

X-25-V single = 223
X-25-V raid0 = 838 ???

That's ~3.75x the iops, i.e. better than 100% scaling....

There must be some other bottleneck in the single drive scenario , that the R0 is alleviating

[alfaunits]

Probably influenced a lot by the cache. Also I think the increased sequential speed in R0 makes the big difference.

That does put weight on Hertz's statement.. i.e. he's right. I was looking at the entire article not the Anand Storage bench only.

[Computurd]

yes there are a few inconsistencies, but overall they do look close. alot of stevero's testing mirrors that of anand, but, not having any of the hardware myself to test, i cannot say definitely how it is.

[SteveRo]

Comparing anand's iometer results to mine - some are close - a few are way off - for example 4KB random reads i get 229MBps (qdepth 64), anand gets 62.6MB/s (aligned) - anand commented that something looked bottlenecked - could also be a difference in qdepth.

That said, I agree with anand's general conclusion and the great idea of leaving spare space to (reduce write amplification) make up for the absence of trim.

I just bot 2x kingston from amazon for $75 each - will convert to intel x25-v's, for $150 bucks hard to beat! - http://www.overclock.net/ssd/656984-...-40gb-ssd.html

[NeedMoMegaHurtZ]

But how is it possible - even theoretically - to get 3.75x the IOPS by adding 1 drive? (again looking at the heavy trace)

What am I missing here?

Even Anand's own sequential read/writes measurements show ~2x the performance , not 3.75x the performance , and the random reads/writes scale much less than 2x.

Even if all workloads were 100% sequential, wouldn't the theoretical maximum be ~2x ? Since there is a mix in the heavy trace, you would expect it to be <2x

I agree with the value in these drives, but one_hertz raised some good questions

[SteveRo]

I can't explain it (other than perhaps the ich10 raid controller or high qdepth or both)- but I was able to confirm ~230MBps random read - using both iometer and AS SSD -

[One_Hertz]

Comparing anand's iometer results to mine - some are close - a few are way off - for example 4KB random reads i get 229MBps (qdepth 64), anand gets 62.6MB/s (aligned) - anand commented that something looked bottlenecked - could also be a difference in qdepth.

That said, I agree with anand's general conclusion and the great idea of leaving spare space to (reduce write amplification) make up for the absence of trim.

I just bot 2x kingston from amazon for $75 each - will convert to intel x25-v's, for $150 bucks hard to beat! - http://www.overclock.net/ssd/656984-...-40gb-ssd.html

I was strictly talking about the scores on the ANAND storage bench. The drives are great, no argument there at all.

[bigretard21]

By looking at those benchmarks, are there any conclusions to be drawn about which drives would perform best as a database/web/file server? I.e., can you look at the heavy workload bench and say the top performers in that category will be the best?

[Spoiler]

By looking at those benchmarks, are there any conclusions to be drawn about which drives would perform best as a database/web/file server? I.e., can you look at the heavy workload bench and say the top performers in that category will be the best?

I hate to draw any conclusion using a Value drive against any database/web/file server usage. If I had to though, I would stick to the drive that delivers better Random Read specs -Intel X25-m or Crucial drive.

[Metroid]

But there is something funky going on with the heavy workload trace...

X-25-V single = 223
X-25-V raid0 = 838 ???

That's ~3.75x the iops, i.e. better than 100% scaling....

There must be some other bottleneck in the single drive scenario , that the R0 is alleviating

The test takes up 30GB to be run, the Intel X25-V has only 37GB. There is no space for this kind test. The space is divided with 2 X25-V 74GB, 15GB each, with spare space to write to it, performance tops up once again.

Anand describes it better,

Our storage bench is based on a ~34GB image, which doesn't leave much room for the 40GB X25-V to keep write amplification under control. With two our total capacity is 74.5GB, which is more than enough for this short workload. With the capacity cap removed, the X25-Vs can scale very well. Not nearly twice the performance of an X25-M G2, but much faster than a single drive from Intel.

Great testing done by Anand Lai Shimpi once again.

[Budwise]

this makes me want to consider selling my X25-M G2 and getting two V's.

[Spoiler]

this makes me want to consider selling my X25-M G2 and getting two V's.

What?!?! Just get another X25-M! Don't go backwards.

But if you do....feel free sell that thing in the 'for-sale' forums. I could always use one.

[Budwise]

well, going to the V's it would only be about $30 out of pocket whereas buying another M G2 would be another $220.

[WC Annihilus]

If you're going to do that, should make sure to check Newegg early tomorrow. Based on the Shell Shocker preview, the V drive should be the earlier one

[SteveRo]

If you're going to do that, should make sure to check Newegg early tomorrow. Based on the Shell Shocker preview, the V drive should be the earlier one

Yes, I saw that too.
Thu am shell shocker - starts when midnight?

[NeedMoMegaHurtZ]

[QUOTE=Metroid;4316971]The test takes up 30GB to be run, the Intel X25-V has only 37GB. There is no space for this kind test. The space is divided with 2 X25-V 74GB, 15GB each, with spare space to write to it, performance tops up once again.

Anand describes it better,

...SNIP...
[QUOTE]

Thanks Metroid, that explains the scaling single drive vs. R0, and was the bottleneck I was looking for. I should have read more closely! DUH!

Cheers

[Biker]

If you're going to do that, should make sure to check Newegg early tomorrow. Based on the Shell Shocker preview, the V drive should be the earlier one

Do you have a link to the weekly deals page?

Wish I lived over there!

If I did i'd be up for 4 if the price was right

[Computurd]

well the anandtech bench holds true. i remember when i was reading steve's thread how impressed i was with the performance of those drives. or the equivalent, whatever.
very impressive. 84 bucks with free shipping? wow.