Sign in with
Sign up | Sign in
Your question

Slow Sequential Writes in the Real World?

Last response: in Storage
Share
January 27, 2010 12:10:40 PM

Dear Board

I have decided to invest in an SSD for my PC. This will act as my system drive and will replace a 300GB Velociraptor. I have, I think, settled on an 80GB Intel X-25M (G2).

From extensive reading around I've concluded that this SSD will be an all round faster performer. However I still don't really understand the implications of the Intel's slow sequential write speed.

Can anyone give me an idea of how this would manifest itself in real world usage:
What type of activities would suffer, if any?
Are there better choices than the Intel? (I have max £200GBP to spend?)
Or should I just stick with the velociraptor?

Thanks in advance
Jonny
a c 127 G Storage
January 27, 2010 8:27:38 PM

SSDs are used mainly for system disks; not as mass storage like 1TB+ HDDs are designed for. On your system disk, you should be seeing close to 90% read 10% write and 90% random I/O and 10% sequential.

Meaning: sequential transfer rates do not affect performance. Sequential speeds are useful for COPY tasks; like copy this 50GB large Music directory from HDD1 to HDD2. For launching applications etc; random I/O is concerned and your SSD will be very fast.

Although the X25-M G2 has been around; its still (one of the) best SSDs available right now. The SandForce controller may give some competition to Intel, though. So its possible you may see new products entering the market soon; i think in March 2010, a new wave of SSDs will hit the market.

But you shouldn't have to worry about that low 75MB/s speed; if you look at IOps performance; the Intel SSDs totally blow away any HDD by many many degrees.
January 27, 2010 8:47:58 PM

Hi sub mesa

Thanks for the reply.

I'm pretty sure I'm ready to take the plunge with the X-25.

Unless you have any better suggestions.

Cheers
Jonny
Related resources
a b å Intel
a c 415 G Storage
January 27, 2010 9:13:50 PM

I have an Intel X25-M G2 as my OS drive. I also have my user home directory there, although I actually keep all my data files on a separate hard drive. With the user home directory on the SSD the profile is stored there, including the myriad of cache, history and classification files used by Firefox (or IE, if you use it instead) - so leaving it there speeds up the launching of my web browser.

I run a program which continuously monitors disk activity and lets me see read and write rates to each drive. The SSD holding the OS gets writes every few seconds even when nothing in particular is happening - but they're very tiny writes taking only a few K each. Even applications like Firefox, which seems to write about 20MB worth of data to the drive when it starts, don't seem to have any lag that I'd really attribute to the disk. So I've found that write performance is really a non-issue for me.

You may want to have a look at a video I made comparing the time needed to boot 64-bit Windows 7 and start Firefox with the OS residing on hard drives vs. the SSD - its at: http://www.youtube.com/watch?v=cTHX0MqVMss
a c 127 G Storage
January 27, 2010 9:14:12 PM

You could wait for new (cheaper) SSDs with SandForce controller. But the price difference wouldn't be all that big, and the Intel controller still does pretty well in real benchmarks of actual workload.

The SandForce controller does perform much better at sequential throughput, though. That would be nice if a 1TB SSD was possible; but with 80 or 160GB you won't have that much sequential access as it often means it'll be the C: system drive, while a larger HDD is the D: data drive to store large files on.

So, yes, the Intel X25-M G2 is still a good choice.
a c 127 G Storage
January 27, 2010 9:17:56 PM

<offtopic>
Hello sminlal, nice to see you're still around here. :) 
a b å Intel
a c 415 G Storage
January 27, 2010 9:27:32 PM

> Hello sminlal, nice to see you're still around here.

I only joined about 6 months ago. I guess that must be, like, 6 decades in Internet time... :D 

Good to see you too!
a c 127 G Storage
January 27, 2010 9:48:25 PM

sminlal said:
> Hello sminlal, nice to see you're still around here.

I only joined about 6 months ago. I guess that must be, like, 6 decades in Internet time... :D 

Good to see you too!

Can hardly believe that! But i guess we posted alot on the same period some months back when i was hyperactive on this forum. :p 

Anyway, i valued your posts and its nice to see you're still around. ;) 
January 28, 2010 7:17:53 PM

Hi Gents

Thanks for the speedy responses and I'm glad to see I've reunited two old posting buddies.

You've helped make my decision, I'll be buying the drive this evening. I have gift vouchers for an online shop and I'm a firm believer in the principle that a gift shouldn't be something you need but something you want. I certainly don't need an SSD but I definitely looking forward to having one.

Thanks again
Jonny
a b G Storage
January 30, 2010 12:10:17 PM

sub mesa said:
SSDs are used mainly for system disks; not as mass storage like 1TB+ HDDs are designed for. On your system disk, you should be seeing close to 90% read 10% write and 90% random I/O and 10% sequential.

Meaning: sequential transfer rates do not affect performance. Sequential speeds are useful for COPY tasks; like copy this 50GB large Music directory from HDD1 to HDD2. For launching applications etc; random I/O is concerned and your SSD will be very fast.

But you shouldn't have to worry about that low 75MB/s speed; if you look at IOps performance; the Intel SSDs totally blow away any HDD by many many degrees.

Without an actual definition of a random or sequential I/O operation it's difficult to make comparisons but let's assume a random I/O is 4 KB and a sequential I/O is 1 MB.

Velociraptor HDD performance:
avg. access time: 7.0 ms
transfer rate: 120 MB/s
write 4 KB: 0.03 ms + 7.0 ms = 7.03 ms
write 1 MB: 8.33 ms + 7.0 ms = 15.33 ms

90 x random I/O = 90 x 7.03 = 632.7 ms = 80% of total time
10 x sequential I/O = 10 x 15.33 = 153.3 ms = 20% of total time
total time: 786 ms

Intel SSD performance:
avg. access time: 0.1 ms
transfer rate: 75 MB/s
write 4 KB: 0.05 ms + 0.1 ms = 0.15 ms
write 1 MB: 13.33 ms + 0.1 ms = 13.43 ms

90 x random I/O = 90 x 0.15 = 13.5 ms = 10% of total time
10 x sequential I/O = 10 x 13.43 = 134.3 ms = 90% of total time
total time: 148 ms

Obviously this is all very simplistic but it shows your statement about the SSD blowing away the HDD is correct. It's more than five times faster.
But your statement about the irrelevance of sequential I/O performance is not correct. For a traditional HDD sequential IOPS take up 20% of the time and for SSD they're actually more important than random IOPS because the SSD spends 90% of its time to perform the 10% sequential IOPS.
a c 127 G Storage
January 30, 2010 3:09:28 PM

With IOPS performance, i mean random I/O IOps; not sequential IOps; sequential performance should just be expressed in MB/s that's the most convenient for many people.

While the SSDs don't excell at sequential throughput, my argument was that a small system drive with many small files won't see much sequential write I/O - but would see alot of random reads and sequential reads; both of these are alot faster than the HDD can achieve.

Also, a velociraptor doesn't pull 120MB/s at the end of its capacity; nearing the inner tracks on the physical platters. Also, filesystem metadata and fragmentation causes even sequential I/O to include some random I/O; which the SSD doesn't notice at all but a few seeks can degrade HDD performance. So with an SSD you're much less likely to notice slow I/O.
a b G Storage
January 30, 2010 10:08:44 PM

There is no clear definition for random I/O IOPs. you can have random IOPs of 4K files and random IOPs of 1MB files. As my example shows this makes a huge difference so I think it's better to use access times and data transfer rates to discuss performance, both of which are easily measurable.

Yes, a SDD is much faster than a HDD but the TS also wanted to know if there where better choices than the Intel. I assume he meant if there where better SSDs for <GBP200.
Random access times are so low with SSDs that you can almost ignore them when comparing SSD performance. So what's important are the data transfer rates which are relevant for both random I/O and sequential I/O.
For HDDs data transfer rates are only relevant for larger data transfers because the random access times are high.

The Velociraptor, like other HDDs indeed gets slower near the inner tracks but this is irrelevant in this case since the TS apparentely only needs the first 80GB.
a b å Intel
a c 415 G Storage
January 30, 2010 11:16:45 PM

Anandtech's review shows that there's a factor of up to 4X difference between various SSDs in terms of 4KByte random I/O performance. That performance is mostly determined by access times, so I think you're mistaken to believe that access times between various SSDs are mostly irrelevant. These differences are way more important than transfer rates for the kind of boot and application load performance that SSDs are best at.




Complete article here: http://www.anandtech.com/storage/showdoc.aspx?i=3631
a b G Storage
January 31, 2010 2:03:15 AM

sminlal said:
Anandtech's review shows that there's a factor of up to 4X difference between various SSDs in terms of 4KByte random I/O performance. That performance is mostly determined by access times, so I think you're mistaken to believe that access times between various SSDs are mostly irrelevant. These differences are way more important than transfer rates for the kind of boot and application load performance that SSDs are best at.

http://images.anandtech.com/reviews/storage/Intel/34nmSSD/Review/used-4kb-write-mbs.png
http://images.anandtech.com/reviews/storage/Intel/34nmSSD/Review/used-4kb-read-mbs.png

Complete article here: http://www.anandtech.com/storage/showdoc.aspx?i=3631

Let's do the math again:

4x slower access times, same transfer rate:
90 x random I/O = 90 x 0.45ms = 40.5ms = 23% of total time
10 x sequential I/O = 10 x 13.83ms = 138.3ms = 77% of total time
total time: 178.8ms which is 21% slower

If you then look at the transfer rates. The review shows up to 2.5x difference between the SSDs, so

2.5x slower transfer rates, same access times:
90 x random I/O = 90 x 0.225 = 20.25ms
10 x sequential I/O = 10 x 33.33 = 333.3 ms
total time: 353.55 ms which is a whopping 140% slower

So even if the access times are 4x higher, they're still too low to make a big difference.
But a lower transfer rate makes a huge difference.
!