Jump to content
Sign in to follow this  
magneto

Is it smart to use 2 M.2 SSDs in raid 0 for temporary caching?

Recommended Posts

Hi,

 

This might be a stupid idea but is it a good idea to use 2 M.2 SSDs in raid 0 for temporary caching?

 

http://www.legitreviews.com/samsung-sm951-m2-pcie-ssds-raid0-performance_161753

 

I don't know if 4.5GB/s is real but if it's, I can only imagine it helping to read big cache files very fast, right? I notice it takes a lot of time to read the cache files of off disks when playing an animation in Houdini, even for 100MB files I think.

Share this post


Link to post
Share on other sites

No, it's not smart in my opinion to ever use RAID 0 for any reason. Though I'm a little biased when comes to data storage leaning towards data integrity and redundancy. Throughput is one factor of many that determines performance. Even if the storage can provide thousands of megabytes a seconds doesn't mean Houdini can process it that quickly. A good test would be to use system memory for a temporary drive and compare that to a single hard drive (best case versus worst case) to get an idea of how much of a factor the storage really is. Then make a decision based on that. My guess is you'll find spending $1,000+ on storage for simulation caching won't be worth it, but a single SSD might be worth it.

  • Like 1

Share this post


Link to post
Share on other sites
Guest tar

I would only Raid 0 to get speed out of cheaper platter disks. SSDs already are quite fast, and with .bgeo.gz/sc cache files being compressed by default it may be CPU bound more than disk speed too.

Edited by tar

Share this post


Link to post
Share on other sites

I was wondering the same thing.  Glad to hear I don't need to buy 2 m.2s for raid 0!

  • Like 1

Share this post


Link to post
Share on other sites

I would use a PCI-Express x4 SSD card if you need very fast caching, such as the Intel 750 series SSD: http://www.newegg.com/Product/Product.aspx?Item=N82E16820167300.

 

Not all RAID controllers support trim in RAID mode. Trim helps keep SSD performance at peak levels and without it performance will degrade over time.

 

Wow, those are NVMe devices? Would Houdini be able keep up with these read speeds? It would be really awesome.

Share this post


Link to post
Share on other sites

Wow, those are NVMe devices? Would Houdini be able keep up with these read speeds? It would be really awesome.

I hope Houdini can keep up!  I want to use one NVMe for os/apps and one for cache/sims i/o.

 

Samsung will release their 950 pro's in October supposedly:  (right now, you have to check the actual part number in order to determine if a SM951 is NVMe or not)!

 

http://www.trustedreviews.com/news/samsung-launches-nvme-m-2-950-pro-ssd-for-the-masses

Share this post


Link to post
Share on other sites

I hope Houdini can keep up!  I want to use one NVMe for os/apps and one for cache/sims i/o.

 

Samsung will release their 950 pro's in October supposedly:  (right now, you have to check the actual part number in order to determine if a SM951 is NVMe or not)!

 

http://www.trustedreviews.com/news/samsung-launches-nvme-m-2-950-pro-ssd-for-the-masses

 

I read for i7 CPUs, using 2 of these with a grapnics card will hinder of the cards' PCI lanes. I will check for samsung too. I thought they only had SM951 which says AHCI.

 

It says it's much faster and the price is cheaper for 512 too. Worth to wait I think unless Mark can chime in :)

http://www.techpowerup.com/216209/samsung-announces-the-950-pro-consumer-m-2-pcie-ssd.html

Share this post


Link to post
Share on other sites
Guest tar

I'm seeing almost no difference between a Ram disk and an SSD in loading ~350MB bgeo/.sc takes ~4secs to load into Houdini.

 

 

Disk speed tests results: Using Black Magic Disk Speed Test:

 

Ram Disk DDR3  1066MHz

2.3 GB/s

 

SSD 'INTEL SSDSC2BP240G4 Media'

0.25 GB/s

 

OSX 10.10.4, H 14.0.453, 2 x 3.33 GHz 6-Core Intel Xeon X5680

Edited by tar

Share this post


Link to post
Share on other sites
Guest tar

We'll have to come up with a good test.

 

Packing some 35MB files, for a 1.5GB sequence, made no difference - thought I could bypass Houdini and display it direct to the GPU via a packed bgeo.

 

A lightweight particle sim was essentially the same read speed @ 27fps.

Edited by tar

Share this post


Link to post
Share on other sites

So no need to jump on these fancy drives for Houdini usage? It would also be interesting to try these ideas in other apps like Maya, but also compositing apps, where I read that they benefit hugely for playback of high res image sequences. 

Share this post


Link to post
Share on other sites
Guest tar

At this stage in testing I would say SSD is good enough, read/write speed may be different though for different things - particles vs vdb etc that is yet to be tested.

 

2d comp apps are similar to Houdini, they have a limit, but 2d editing apps can really benefit as they don't need to cross reference other pixels in general AFAIK. 

 

Edit: A few more tests and thus far it appears on this computer Houdini can load and display at ~100MB/sec

Edited by tar

Share this post


Link to post
Share on other sites

While manufacturers tout the sequential read and write speeds of SSDs, where you see the real difference between an SSD and a HD RAID0 setup is random read and write speed. Random reads occur when a lot of small files are accessed, or a few large files are randomly seeked (ie, Alembic, RAT). HD performance falls through the floor in this case, handling barely 1MB/s in the 4K random read test.  SSDs can often do almost 100MB/s in this test, especially the high-end ones.

 

But before dropping a lot of money on a big SSD for caching, try a small SSD (if you have one around) and see if disk access is a bottleneck for you. I have a 256GB system SSD, and a small 64GB SSD that I use for my swap partition and /tmp (really nice when my mem gets full, the system doesn't grind to a screeching halt as it swaps). The HD still does all the heavy lifting of test cases & data. I've found that yes, the SSD does load some scenes a bit faster, and is faster for Alembic and image reading, but not so significantly so that I'd pony up for a 1TB SSD.

Share this post


Link to post
Share on other sites

While manufacturers tout the sequential read and write speeds of SSDs, where you see the real difference between an SSD and a HD RAID0 setup is random read and write speed. Random reads occur when a lot of small files are accessed, or a few large files are randomly seeked (ie, Alembic, RAT). HD performance falls through the floor in this case, handling barely 1MB/s in the 4K random read test.  SSDs can often do almost 100MB/s in this test, especially the high-end ones.

 

Another factor is what the application does with the data. Does it have to decompress it, does it have to reorganize or process it, does it have to cull it, etc. The processing of the data before the user can do something with it might take far longer than any disk IO.

Share this post


Link to post
Share on other sites

While manufacturers tout the sequential read and write speeds of SSDs, where you see the real difference between an SSD and a HD RAID0 setup is random read and write speed. Random reads occur when a lot of small files are accessed, or a few large files are randomly seeked (ie, Alembic, RAT). HD performance falls through the floor in this case, handling barely 1MB/s in the 4K random read test.  SSDs can often do almost 100MB/s in this test, especially the high-end ones.

 

But before dropping a lot of money on a big SSD for caching, try a small SSD (if you have one around) and see if disk access is a bottleneck for you. I have a 256GB system SSD, and a small 64GB SSD that I use for my swap partition and /tmp (really nice when my mem gets full, the system doesn't grind to a screeching halt as it swaps). The HD still does all the heavy lifting of test cases & data. I've found that yes, the SSD does load some scenes a bit faster, and is faster for Alembic and image reading, but not so significantly so that I'd pony up for a 1TB SSD.

 

I read that Samsung 950 Pro is gonna have better random read than Intel 750 that focuses on sequential I think.

Share this post


Link to post
Share on other sites
Guest tar

For SSDs I didn't know how accurate the reports that 'trim' was important but OsX has shown it is. ~8 months of usage has dropped the  Write speed from 230MB/s to ~110MB/s. Read speed is the same at ~265MB/s.

 

Luckily the latest OsX now support it for non-apple branded disks.

Share this post


Link to post
Share on other sites
Guest tar

While manufacturers tout the sequential read and write speeds of SSDs, where you see the real difference between an SSD and a HD RAID0 setup is random read and write speed. Random reads occur when a lot of small files are accessed, or a few large files are randomly seeked (ie, Alembic, RAT). HD performance falls through the floor in this case, handling barely 1MB/s in the 4K random read test.  SSDs can often do almost 100MB/s in this test, especially the high-end ones.

 

 

This pretty much bears out in a recent test:

 

LaCie 4Big striped as Raid 5+1(6), eSata card to Mac Pro Cheesegratter. OsX 10.10.5

 

Read: 214 MB/s

Write: 227 MB/s

Rewrite 67.9 MB/s

 

Read latency: 427 ms

Write latency: 112ms

Rewrite latency: 436 ms

 

Intel SSD  sata 3Gb/s connection. Trim just enabled.

 

Read: 269 MB/s

Write: 91.3 MB/s

Rewrite 89.4 MB/s

 

Read latency: 13 ms

Write latency: 154 ms

Rewrite latency: 86 ms

Edited by tar

Share this post


Link to post
Share on other sites

http://www.computerworld.com/article/2987956/solid-state-drives/intel-samsung-launch-their-fastest-ssds-with-up-to-5gbps-speeds.html

 

intel.web.480.270-100618367-large.idge.p

 

"Intel's new P3608 Series PCIe SSD comes in 1.6TB, 3.2TB and 4GTB capacities and delivers up to 850,000 I/Os per second (IOPS) random reads and 150,000 random writes.

 

The drive has a sequential read/write speed of up to 5,000MBps (5Gbps) and 3,000MBps (3Gbps), respectively.

 

Intel did not offer pricing for either of its new SSDs."

Share this post


Link to post
Share on other sites
Guest tar

 

The drive has a sequential read/write speed of up to 5,000MBps (5Gbps) and 3,000MBps (3Gbps), respectively.

 

 

 

That's not amazing though - in MegaBytes it's read 625, write 375

 

EDIT: Oh I see - the Bytes and Bits are mixed up. Not sure what the real data is. One is amazing, the other is okay.

Edited by tar

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

×