Category Archives: Uncategorized

San Testing round 1

Decided it was time to do some testing on my sans to get an idea of what iops they are capable off. I used a windows 2008 r2 base server to start my testing, with the MS mpio drivers. Test were ran using the sqlio tool from microsoft. I will add the vendor mpio drivers later. I am testing the EMC VNX, NetApp FAS3210 , IBM DS8100 behind IBM SVC and VNX behind IBM SVC. SVC and the VNX are on the 8Gbs FC, while the FAS3210 and IBM DS8100 are on 4gbs FC.

The VNX is using auto tier (FAST VP), wtih 200 GB of FAST Cache, the Raid 5 storage pool that has 100 gb SSD, 600gb and 300gb 15K, 300gb 10K and 2tb 7.2K drives in it.

Netapp FAS3210 I am using an Aggregate with 15 1tb 7.2 k disk.

DS8100 is a Raid 10 with 96 73gb 15k disks.

The first round of test were done with w2k8 r2 right out of the box using the windows native mpio driver. SQLio is using a 64k block size and writing or reading a 50gb file. Here are the Results. As expected the 8100 was the best performer, the NetApp and the VNX were Close, in fact the Netapp beat out the VNX .

Storage System iOs/Sec Mbs/Sec Ran/SEQ io Read or Write
8100 4700.85 293.8 Random Write
8100 4820.72 301.29 Sequential Write
8100 15115.02 944.68 Random Read
8100 13950.89 871.93 Sequential Read
VNX/Auto Tier 445.94 27.87 Random Write
VNX/Auto Tier 3783.15 236.44 Sequential Write
VNX/Auto Tier 4007.63 250.47 Sequential Read
VNX/Auto Tier 1570.34 98.14 Random Read
NETAPP 883.45 55.21 Random Read
NETAPP 284.96 17.81 Random Write
NETAPP 4503.39 281.46 Sequential Read
NETAPP 2725.82 170.36 Sequential Write
VNX/SVC 4638.99 289.93 Sequential Write

The next set of test I loaded SP1 for w2k8, I know there were some improvements in the mpio driver, however… I was not expecting the results I got below.  The VNX and the 8100 made huge improvements on write where reads were a little slower with the new mpio driver. However the FAS3210 had a decrease in io across the board. Later on I will load the Vendor MPIO drivers and see what the results are.

Storage System iOs/Sec Mbs/Sec Ran/SEQ io Read or Write
8100 7327.36 457.96 Random Write
8100 7439.3 464.95 Sequential Write
8100 10815.27 675.95 Random Read
8100 16070.83 1000.42 Sequential Read
VNX/Auto Tier 7372.79 460.79 Random Write
VNX/Auto Tier 7168.43 448.02 Sequential Write
VNX/Auto Tier 3001.21 187.57 Sequential Read
VNX/Auto Tier 6668.09 416.75 Random Read
NETAPP 681.34 42.58 Random Read
NETAPP 693 43.31 Random Write
NETAPP 2147.03 134.18 Sequential Read
NETAPP 1378.91 86.18 Sequential Write
VNX/SVC 4593.66 287.1 Sequential Write
VNX/SVC 632.38 39.52 random Write

More to come later I am going to tweak the block size and see if I can find were they perform the best.


Had a chance to go to an EMC XtremIo briefing this week. XtremIO is EMC’ss entry into the all flash array vertical. The presenter made and interesting point right off the start, which was that it as not about the flash in the array, but about the software. XtremIO takes all the data coming in and hash’s it into 4k bits, and the writes data to flash when it has a full stripe. It also does in-line de-duplication. The performance numbers for one X brick are 150,000 iops with random reads and writes on an array that is 50% full. These numbers were also given when the array is getting data that is greater than 4k blocks, so these appear to be real world numbers. Performance will go up (essentially double with each x brick) as you add X bricks, right now the max  will be 8 X bricks, but the software has been designed where there is no limit on the amount of x bricks that could be in a XtremIO flash array. The files system is not a traditional raid it is n+2 and was built specifically to handle an all flash array. Native replication for this array will be out next year and it will only offer asynchronous replication , as synchronous would affect the overall performance of the array. If you need replication now you can use vplex, and native support for Recover point will be coming and the end of this year.

I like the scalability, the fact you can use all off the array (some vendors recommend not to use more than 80% capacity on their flash array’s). They only other flash array I can compare this too is the IBM flash systems with IBM seems to recommend deploying behind SVC , which I understand since the FlashSystems are not scalable and do not have any replication natively.  With XtremIO you will not need a second device for replication and if you need more space you can just add another x brick and get more performance at the same time. XremeIO looks very interesting to me and it will be interesting to see how it is adopted in the market place and what other Flash Arrays are coming from other vendors.