2

I recently purchased two HP 380P Gen 8 servers with 365GB IODrive2s for VDI using Vmware View 5.2.

After using the latest HP vmware 5.1 update 1 ESX installer and then installing the Fusion-io drivers for ESX I was able to format the drive as VMFS5, create a pool and put the replica on the Fusion-io.

All of these seemed to run about as fast as our two 15k SAS drives in RAID0. Unsatisfied with the VDI experience we decide to create a pool with all components on the Fusion-io. The speed still didn't seem on par with what the spec sheets state. I finally did a test copy of from the Fusion-io to itself (from the ESXi ssh shell) and I am currently getting about 1GB per minute transfer rate. The white papers claim I should be getting around 500MBps.

Here are some tests:

# ls -lrt
rw-------    1 root     root     3572297728 Aug 22 11:43 VMware-VIMSetup-all-5.1.0-947939.iso
# du -h
3.3G    .

from 4 drive RAID10 15k SAS to itself:
cp VMware-VIMSetup-all-5.1.0-947939.iso test.iso takes 37s

from 4 drive RAID10 15k SAS to Fusion-io:
cp VMware-VIMSetup-all-5.1.0-947939.iso /vmfs/volumes/fusion2/test.iso takes 34 sec

from Fusion-io to itself:
cp test.iso test2.iso takes 78 sec

from my desktop SSD to itself: 32s

7200rpm hard drive to itself: 115s

I currently have a ticket open with HP and VMware but I am essentially getting fingerprinting. I am hoping someone in the community has solved my issues.

Outputs from IOMeter using 8 workers with the all in one access spec for 1 hour:

Samsung810 SSD
IOPS        Read IOps       Write IOps      MBps            Read MBps       Write MBps  Transactions per Second
3385.116269 1692.255704 1692.860565 43.364704   21.683055   21.681649   3385.116269

Fusion-io:
IOPS        Read IOps       Write IOps      MBps            Read MBps       Write MBps  Transactions per Second
13172.96404 6570.691766     6602.272278     168.838142      84.220818       84.617323   13172.96404

4disk15kSAS:
IOPS        Read IOps       Write IOps      MBps            Read MBps       Write MBps  Transactions per Second
62854.07353 31429.66595 31424.40759 805.541003  402.802938  402.738065  62854.07353
Improve this question
8
  • 1
    Moar details! What was wrong with the performance?
    – ewwhite
    Aug 22, 2013 at 17:52
  • Throughput is 1 GB per minute. This is an unused server I know I may not reach the theoretical 500MB/sec (30GB/min). But 1GB/min is definitively not good. I have tried copying to the 15K SAS raid-10 datastore to the fusionio and fusionio to itself with the same results.
    – Bernie
    Aug 22, 2013 at 18:44
  • Sequential I/O throughout isn't the purpose or best measure of an SSD. Latency and random read/write performance is. Can you try a more realistic use-case and see if things look any better?
    – ewwhite
    Aug 22, 2013 at 22:48
  • I am currently running iometer to get more concrete numbers.
    – Bernie
    Aug 23, 2013 at 16:25
  • I haven't used FusionIO with VMWare before so I don't really have any answers for you.. but the performance is definitely not anywhere close to what you should be getting. Here's what I get creating a 1GB file on FusionIO under CentOS 6. # time (dd if=/dev/zero of=1gigfile bs=1M count=1024 && sync) 1024+0 records in 1024+0 records out 1073741824 bytes (1.1 GB) copied, 0.686625 s, 1.6 GB/s real 0m1.107s user 0m0.000s sys 0m0.689s So over 1.6GB/sec written, 1.1 seconds to write and sync.
    – user143703
    Aug 23, 2013 at 19:15

1 Answer 1

Reset to default
0

Everything is working now. Thanks for all of you help. I wasn't testing the card vs the sas array hard enough. Once I used random writes and reads in testing and saturated the cache on the raid controller, the fusionio beat the array. 168MB/s vs 6 MB/s. I was also able to speed everything up in VDI by enabling a portion of the fusion to be used as host cache.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .