Hello everybody,
the old thread seems to be sooooo looooong - therefore I decided (after a discussion with our moderator oreeh - thanks Oliver -) to start a new thread here.
Oliver will make a few links between the old and the new one and then he will close the old thread.
Thanks for joining in.
Reg
Christian
I've updated my numbers above for the EMC Celerra NS-120 swISCSI and NFS tests. EMC best practices were applied to improve NFS performance. EMC also made the recommendation that the IOMETER test script should include 120 seconds of Ramp Up time. The current Ramp Up time of 0 seconds can skew test results. My new numbers include the Ramp Up time of 120 seconds for the swISCSI and NFS tests. I haven't retested fibre channel.
[i]Jason Boche, vExpert[/i]
[boche.net - VMware Virtualization Evangelist|http://www.boche.net/blog/][/i]
[VMware Communities User Moderator|http://www.vmware.com/communities/content/community_terms/][/i]
[Minneapolis Area VMware User Group Leader|http://communities.vmware.com/community/vmug/us-central/minneapolis][/i]
[vCalendar|http://www.boche.net/blog/index.php/vcalendar/] Author[/i]
Here is what I got. We were looking for a low cost storage sans to dump some exchange databases on. This unit from Promise seemed to fit the bill.
We've had it for 3 months with no issues. MIght get another for storage at out DR site.
ANY SUGGESTIONS ON HOW I MIGHT IMPROVE THINGS IS WELCOMED.
SERVER TYPE: VMWare ESXi 4.
CPU TYPE / NUMBER: VCPU / 1
HOST TYPE: HP DL360 G5, 12GB RAM, 1 x Intel E5440, 2.83GHz, QuadCore
STORAGE TYPE / DISK NUMBER / RAID LEVEL: Promise VessRaid 1830i (512MB CACHE/SP) / 4 SATA WD 300GB Velociraptor 10k/ R10 -
SAN TYPE / HBAs : Ethernet 1Gb; VMWare iSCSI software initiator (HP NC360T)
##################################################################################
BASE Test--
##################################################################################
Max Throughput-100%Read........___17.1298_____.......___3497.61_____.....____109.30______
RealLife-60%Rand-65%Read......_____71.1607____.....____750.30____.....____5.86______
Max Throughput-50%Read.........____15.9227____.....____3812.36____.....____119.14______
Random-8k-70%Read..............____74.2_____.....____722.45______.....____5.64______
Jumbo Frames and Flow Controll on Stroage VLAN - HP 1810G 24port Switch
SERVER TYPE: VM
CPU TYPE / NUMBER: VCPU / 4
HOST TYPE: DL380 G5, 32GB RAM; 2 x Intel E5440, 2.83GHz, QuadCore
STORAGE TYPE / DISK NUMBER / RAID LEVEL: PS4000E / 14+2 DISK (7.2K SATA) / R50)
NOTES: 2 NIC, MS iSCSI, no-jumbo, flowcontrol on
##################################################################################
TEST NAME--
##################################################################################
Max Throughput-100%Read........___17.92____..........___3156____.........___98.6____
RealLife-60%Rand-65%Read.......___19.33____..........___2176____.........___17.00__
Max Throughput-50%Read.........___13.86____..........___3828____.........___119____
Random-8k-70%Read..............___22.15____..........___1943____.........___15.19__
EXCEPTIONS: CPU Util.-XX%;
##################################################################################
I believe the throughput is limited by the NIC configuration, otherwise the numbers look as they should for this configuration.
SERVER TYPE: VM XP SP3 on ESX3.5 U4
CPU TYPE / NUMBER / RAM: vCPU / 1 / 1GB
HOST TYPE: Dell PowerEdge 2950, 24GB RAM, 2x Intel X5450 3.0Ghz
STORAGE TYPE / DISK NUMBER / RAID LEVEL: NetApp FAS3020c / 14x 300GB 10k 2Gb FC (per loop) / RAID DP
SAN TYPE / HBAs: iSCSI / sw iSCSI
OTHER: Disk.SchedNumReqOutstanding and HBA queue depth set to 64 on ESX host
Note that the LUN provisioned to ESX spans over one aggregate on one loop, i.e. 14 disks are used
Moreover, NO jumbo frames were used
Test Name | Avg. Response Time | Avg. I/O per Second | Avg. MB per Second |
Max Throughput - 100% Read | 17 | 3385 | 106 |
Real Life - 60% Rand / 65% Read | 30 | 1609 | 13 |
Max Throughput - 50% Read | 13 | 4020 | 126 |
Random 8K - 70% Read | 32 | 1434 | 11 |
Are those good values?
Ich bin vom 8.2. bis 10.2.2010 nicht im Büro und kann deshalb keine E-Mails lesen. E-Mails werden nicht weitergeleitet.
In dringenden Fällen wenden Sie sich bitte an meine Kollegen unter 0351 / 49701-150 bzw. per E-Mail saxonia.hotline@saxsys.de.
SERVER TYPE: VMWare ESX 4u1
GUEST OS / CPU / RAM Win2K3 SP2, 1 VCPU, 2GB
HOST TYPE: IBM HS21 XM, 32GB RAM, 2 x Intel E5420, 2.50GHz, QuadCore
STORAGE TYPE / DISK NUMBER / RAID LEVEL: IBM Storage Module, 6 drive 300GB 10K, RAID 10
SAN TYPE / HBAs : SAS Shared
NOTES: Bladecenter S
Access Specification Name | Avg. Response Time | Avg. I/O per Second | Avg. MB per Second | CPU Utilization |
---|---|---|---|---|
Max Throughput-100%Read | 9.02 | 5,613.22 | 175.41 | 25.17 |
RealLife-60%Rand-65%Read | 49.70 | 1,073.48 | 8.39 | 15.06 |
Max Throughput-50%Read | 7.84 | 5,855.29 | 182.98 | 31.41 |
Random-8k-70%Read | 29.28 | 1,788.42 | 13.97 | 17.33 |
Any comments?
Reg
Dario
SERVER TYPE: VM 2003 Server ESX4
CPU TYPE / NUMBER / RAM: vCPU / 2 / 1GB
HOST TYPE: HP DL370 G6, 24GB RAM, 2x Intel X5430 2.4Ghz
STORAGE TYPE / DISK NUMBER / RAID LEVEL: DAS 15x 146GB SCSI 10K in RAID6
During this test I had 4 VMs running these tests at the same time. Includig the below test of Server 2008 R2 which had very poor Real Life and Random tests compared to this 2003 VM. The Max Throughput tests were very comparable.
Test Name | Avg. Response Time | Avg. I/O per Second | Avg. MB per Second |
Max Throughput - 100% Read | 10.1 | 5812 | 181.6 |
Real Life - 60% Rand / 65% Read | 2.6 | 20452 | 159.7 |
Max Throughput - 50% Read | 8.8 | 6631 | 207.2 |
Random 8K - 70% Read | 2.6 | 21030 | 164.3 |
SERVER TYPE: VM 2008 Server R2 ESX4
CPU TYPE / NUMBER / RAM: vCPU / 2 / 1GB
HOST TYPE: HP DL370 G6, 24GB RAM, 2x Intel X5430 2.4Ghz
STORAGE TYPE / DISK NUMBER / RAID LEVEL: DAS 15x 146GB SCSI 10K in RAID6
I thought this was very odd that the results were so poor for this VM in the RealLife and Random tests. Any idea why these tests are so poor in 2008?
Also, during this test I had 4 VMs running these tests at the same time.
Test Name | Avg. Response Time | Avg. I/O per Second | Avg. MB per Second |
Max Throughput - 100% Read | 2.3 | 12450 | 393 |
Real Life - 60% Rand / 65% Read | 32.9 | 243 | 1.9 |
Max Throughput - 50% Read | 8.5 | 6546 | 204.5 |
Random 8K - 70% Read | 24.9 | 157 | 1.2 |
Hi,
Your numbers look too good to be true.
If I read your results correctly - DAS Random 8K test results at 21000 IOPS for only 15 spindles in RAID6? Simply impossible for "rotating rust".
If those drives were SSDs - my apology
What was the size of your test io file?
You are probably testing against cache only, or your results are affected by high cpu usage, or other issue.
Regards,
Radim
They are 10k SCSI disks. And yes, I was using the 8k Random test from the config file. It could be controller or hard drive cache, not sure, also the CPU while running the test was around 50% for 2 cores. I too thought the scores were very high comparing to other results posted...
-Chad
Try to expand test file size to at least 2x size of the cotroller cache and check its size prior tests.
Delete the original test file prior launching iometer and change sector count to apropriate number.
The controller is 256MB, where do I set the test file size? Is it the transfer request size? Also found it is 36MB read cache and 108MB write cache.
Also, I'm now quite certain the high IOPS is due to the controller cache, I migrated my test VM to another array of 4x 500GB 7.2k Sata Raid5 and it had 20,000+ IOPS albeit with a lower MB/s.
-Chad
SERVER TYPE: VMWare ESX 4u1
GUEST OS / CPU / RAM Win2K3 SP2, 2 VCPU, 2GB
HOST TYPE: DELL R610, 32GB RAM, 2 x Intel E5520, 2.27GHz, QuadCore
STORAGE TYPE / DISK NUMBER / RAID LEVEL: PILLAR DATA AX500 180 drives 525GB SATA, RAID5
SAN TYPE / HBAs : FCOE CNA EMULEX LP21002C on NEXUS 5010
TEST NAME--
##################################################################
Max Throughput-100%Read....5.1609..........11275......... 362.86 CPU=22.84%
RealLife-60%Rand-65%Read...3.2424......... 17037........ 131.68 CPU=32.6%
Max Throughput-50%Read......4.2503 .........12742 ........ 403.35 CPU=26.45%
Random-8k-70%Read.............3.2759..........16824.........128.19 CPU=30.39%
##################################################################
The SAN is also running 214 other VMs ...
I have been using iometer for a while now to get some data for my SAN. I created a test vm and have been migrating it to each san to get results. The interesting thing is that the first time that ran IOmeter the cpu in the program spiked and stayed at 100% and I was getting unbeleivable results. I have since moved that vm to another lun and the cpu is arround 40% with good not greeat results. Now when I move it back to tht lun that was giving me 100% it is not happening any longer. Any ideas or thoughts?
I'm not 100% sure, but it sounds similar to some results I was having posted just above. It was when I had the disk controller cache on I was getting amazing results (20,000+ IOPs) and was running ~50% CPU. When I turned the disk controller cache off I got very pedestrian results and CPU went down to 5-10%. Not sure if that relates, but sounds similar to what I encountered.
-Chad
Sorry to ask this again, but... Can someone please help me with these questions so I can make sure I'm running IOmeter properly??
Thanks in advance.
J
Sorry for the ignorance. I am very new to IOMETER and I'm really trying to understand how to run these tests. Few questions...
1. When running these tests from inside a VM, should I be running them on a yellow icon (the VMs drive c: drive itself), or should I be creating a new unformatted drive that will show up in blue? Are people standardizing here?
2. Should I be using the .ICF file from the "original" storage performance thread?
3. When running these tests, are people running them on LUNs in production with VMs on them? Off-hours so disk activity is minimal? For example, my ESX box is connected to a iSCSI SAN with a single LUN/'Datastore on it that has 10 VMs. Should I just run the tests during normal hours? Or are people shutting down all VMs on the Datastore before running these, etc.??
4. I am really confused about the results people are posting. Everyone lists a Av. IOs/sek column, and a Av. MB/sek column, but I don't even see those in my csv results file (??). Are those the same as Total I/O's per second and Total MB's per second in my spreadsheet? If not, where am I supposed to look?
5. Which of the 4 tests are the best idea for how overall VM performance would be?
Sorry for the ignorance.
Greetings again...
We have a Promise M610i in our data center that's mainly used for D2D or D2T stuff. Past attempts to have it interact with VMWare have gone pretty horribly. However, Promise recently released a new firmware that's actually certified under VMWare... and when we did an upgrade on the array to 2G disks, I decided to re-run the battery of tests on it.
Following is the test results for various RAID levels. Each array was created and only had one VM running. I'm using ESX 4's iSCSI initiator in each test, with 2 NICs setup with MPIO. The Promise box has 2x 1G ports on it, each of which are running LACP back to an HP 2848.
Here are the results from a 1MB stripe on the array - 15 2TB WD RE3 7.2K SATA disks (the 16th is a hot spare). The arrays are created by the management software, so I can only assume each array spreads across all 15 disks.
The numbers are actually quite impressive... this setup will cost you about $10K for 32TB of raw storage. If I have time, I might try running 8 tests at the same time against the device and see if I can break it
Also I should have the same set of results soon from a 64K stripe... I was curious myself as to which would be best for VMWare. I might even do some MS iSCSI Initator tests as well - we'll see.
I find it very interesting how much better RAID6 is vs. a RAID 5 or 50 config on this box - not what I expected to see.
Hopefully this formats ok:
RAID0 - 1M Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.698648 | 3,552.02 | 111.000492 | 42.71 |
RealLife-60%Rand-65%Read | 17.197776 | 2,999.36 | 23.432511 | 47.47 |
Max Throughput-50%Read | 9.772605 | 6,075.89 | 189.871513 | 52.00 |
Random-8k-70%Read. | 20.884025 | 2,480.22 | 19.376709 | 44.59 |
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID1E - 1M Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.703526 | 3,547.56 | 110.86137 | 42.70 |
RealLife-60%Rand-65%Read | 24.452243 | 2,006.67 | 15.677145 | 45.88 |
Max Throughput-50%Read | 11.085448 | 5,347.37 | 167.105412 | 47.62 |
Random-8k-70%Read. | 25.384678 | 2,027.33 | 15.83852 | 41.71 |
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID5 - 1M Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.7175 | 3,547.87 | 110.87082 | 43.07 |
RealLife-60%Rand-65%Read | 25.47476 | 1,894.31 | 14.79928 | 45.16 |
Max Throughput-50%Read | 23.764235 | 2,101.26 | 65.664415 | 43.51 |
Random-8k-70%Read. | 31.122049 | 1,580.53 | 12.347879 | 42.12 |
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID50 - 1M Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.670398 | 3,553.60 | 111.050151 | 43.41 |
RealLife-60%Rand-65%Read | 25.641059 | 1,890.47 | 14.769261 | 44.43 |
Max Throughput-50%Read | 27.965077 | 1,999.05 | 62.470316 | 33.63 |
Random-8k-70%Read. | 31.866895 | 1,560.48 | 12.191244 | 41.90 |
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID6 - 1M Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.712874 | 3,548.61 | 110.894147 | 42.83 |
RealLife-60%Rand-65%Read | 24.341849 | 2,010.24 | 15.704977 | 45.89 |
Max Throughput-50%Read | 11.128691 | 5,336.44 | 166.76372 | 47.52 |
Random-8k-70%Read. | 25.380012 | 2,014.46 | 15.738006 | 43.14 |
-
Carter Manucy
Ich bin vom 22. bis 26.3.2010 nicht im Büro und kann E-Mails nur eingeschränkt bearbeiten. E-Mails werden nicht weitergeleitet.
In dringenden Fällen wenden Sie sich bitte an meine Kollegen unter 0351 / 49701-150 bzw. per E-Mail saxonia.hotline@saxsys.de.
Here are the test results from a 64K Stripe (vs 1MB above). I note that 1M seems the way to go with this box, however, RAID5/50 has a real issue with 1M stripes apparently. I re-rant the tests three times and got similar results - for some reason the Promise box gives less that 50% I/O on the 50% Read/Write test.
I also note after running these tests there are only two logical choices for a RAID set on this box - you either go with RAID0 or RAID6.
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID 0 - 64K Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.694244 | 3,548.69 | 110.896489 | 43.33 |
RealLife-60%Rand-65%Read | 19.801217 | 2,636.31 | 20.596148 | 44.16 |
Max Throughput-50%Read | 9.854979 | 5,998.85 | 187.464111 | 52.06 |
Random-8k-70%Read. | 22.863659 | 2,290.94 | 17.897995 | 42.30 |
Frontend: IBM x3650, 24GB, 2 Xeon 5450's, 2 Teamed NetExtreme BCM5704 on HP 2824 w/MPLS, ESX 4.0, WinXP VM, VMWare iSCSI Initiator |
|
RAID1E - 64K Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.703972 | 3,545.64 | 110.801323 | 45.42 |
RealLife-60%Rand-65%Read | 27.012952 | 1,810.03 | 14.140862 | 45.44 |
Max Throughput-50%Read | 11.1241 | 5,332.29 | 166.634217 | 47.40 |
Random-8k-70%Read. | 27.122818 | 1,876.15 | 14.657388 | 42.16 |
Frontend: HP DL385, 12GB RAM, 2xOpteron 2.6 DC, 2xIntel e1000 PCI-X teamed NICs, VMWare ESX 3.0.2, Win2K3 SP2 VM, 1MB BS, MS iSCSI Initiator |
|
RAID5 - 64K Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.695206 | 3,548.65 | 110.895285 | 43.04 |
RealLife-60%Rand-65%Read | 29.524795 | 1,664.85 | 13.006644 | 42.43 |
Max Throughput-50%Read | 11.629027 | 4,336.17 | 135.505391 | 55.51 |
Random-8k-70%Read. | 34.963324 | 1,441.83 | 11.264297 | 39.43 |
Frontend: HP DL385, 12GB RAM, 2xOpteron 2.6 DC, 2xIntel e1000 PCI-X teamed NICs, VMWare ESX 3.0.2, Win2K3 SP2 VM, 1MB BS, MS iSCSI Initiator |
|
RAID50 - 64K Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.69207 | 3,549.42 | 110.919421 | 43.27 |
RealLife-60%Rand-65%Read | 31.85733 | 1,482.93 | 11.585355 | 44.82 |
Max Throughput-50%Read | 12.067723 | 4,540.16 | 141.880088 | 49.17 |
Random-8k-70%Read. | 36.836303 | 1,321.80 | 10.326583 | 41.57 |
Frontend: HP DL385, 12GB RAM, 2xOpteron 2.6 DC, 2xIntel e1000 PCI-X teamed NICs, VMWare ESX 3.0.2, Win2K3 SP2 VM, 1MB BS, MS iSCSI Initiator |
|
RAID6 - 64K Stripe | ||||
TEST NAME | Av. Resp. Time ms | Av. IOs/sek | Av. MB/sek | CPU Use |
Max Throughput-100%Read. | 16.697074 | 3,545.16 | 110.786351 | 43.39 |
RealLife-60%Rand-65%Read | 35.59834 | 1,292.98 | 10.101445 | 45.97 |
Max Throughput-50%Read | 13.34382 | 3,775.64 | 117.988732 | 52.08 |
Random-8k-70%Read. | 41.282322 | 1,155.31 | 9.025888 | 42.89 |
-
Carter Manucy
-