Sign in or 
Location: Ask the Storage Experts
Discussion: MD3000i and write perf
Category:
Discussion Forum /
General Discussion
Share this
Showing 5 posts
|
petrak@monogram.sk |
MD3000i and write perf
Feb 6 2010, 10:19 AM EST
We are experiencing problem with write performance on MD3000i over iSCSI from VMWare2 SATA disks in raid 0 with write perf about 30MB/s 2 SAS disks in raid 0 with perf about 70MB/s MD3000i firmware is the newest one ( 07.35.31.60 ) No disk firmwares available Switch , VMWare and MD3000i set to use jumbo frames With other types of raid write perfs are even worse Read performance is around 110MB/s which is excelent but the writes ...... Are this numbers ok ? Thanks Do you find this valuable?
Keyword tags:
best practice
community
design
experts
storage
|
KongY@Dell |
1. RE: MD3000i and write perf
Feb 11 2010, 12:02 PM EST
| Post edited: Feb 11 2010, 12:03 PM EST
petrak,What's your IO profile- IO size (4k, 8k, 16k, etc)? Is it sequential or random IO? Is this your actual workload or are you using a tool such as IOMeter? Please note that on reads depending on application and even certain IO stress tools, memory caching can take place. Therefore, you may see higher Read performance than Write performance. One method is to look at the packets/throughput at your switch to see if (in = out). You can also compare that throughput with your esxtop counters and perfmon (if you're using Windows VMs) / iostat (if you're using Linux VMs) esp disk throughput. Do you find this valuable? |
|
MiroPetrak |
2. RE: MD3000i and write perf
Feb 12 2010, 1:05 PM EST
those numbers are from bonnie++ and diskbench when performing sequential write ( set to create several 1GB files) , there's no workload at this time, i'm performing tests to know what i can expect from MD3000i. I've checked the utilization of switch port during test and during the read the utilization was about 80 ~ 90% so i think cache didn't take place. Strange for me is that when i have RAID 1 with 2 SATA disks directly on PowerEdge server with PERC6 write perf of this config is ~70MB/s but when i try RAID 1 with 2 SATA disks on MD3000i write perf goes down to ~15MB/s !!! That's what i don't understand ... such degradation of speed Do you find this valuable? |
|
Budrumi |
3. RE: MD3000i and write perf
Mar 23 2010, 7:50 AM EDT
Hi.We've faced similar performance problem with MD3000i. Using SATA disks, the raw sequential write speed never goes above 30MB/s (occasionally up to 40). When you do a multi-threaded seq. write test, you'll notice, the performance degrades even further. We've concluded these tests for example using 4 virtual machines on ESXi that had their primary system partition as either RAW LUN or as virtual disk on VMFS LUN. With 4 concurrent seq. writes (to different LUNs/virtual disks) you might get as low as <10MB/s per VM, total never exceeds 40MB/s. Please note that the test were done on a 14-disk RAID10 array, so the disks were hardly a bottleneck although being SATA disks and not SAS. We've consulted this extensively with local Dell and the end result was that this is a "feature" of this array and that for seq. writes, we should never expect anything much higher than this. The reasoning was that MD3000i is a entry-level solution. Btw., the limitation is on those two RAID controllers in the MD3000i, nothing to do with network setup or disks used. Oh btw., don't ever buy the premium features for MD3000i - snapshots and virtual disk copy, they're pretty much useless (after some insisting, this was also confirmed by Dell) - they take unreasonable ammount of time and hugely degrade array performance while they're performed. However your results got me curious - what kind of SAS disks have you used? 15k RPM? I can't recall, but colleague of mine swears we've also done tests with SAS disks (don't know what type either, ie. 7k, 10k or 15k RPM) and he claims the performance was better than with SATA disks, but only marginally. Do you find this valuable? |
|
MiroPetrak |
4. RE: MD3000i and write perf
Mar 23 2010, 9:43 AM EDT
I've had exactly same answer from local dell support that it is entry-level solution and you can't expect more. But i wasn't satisfied with that answer and I've searched further and found this http://en.community.dell.com/forums/p/19279005/19657765.aspx#19657765.I've contacted customer care and they replaced our SATA disks with NearLine SAS disks. Voila! ..... that was the problem .... NearLineSAS disks have the same write perf as 15k SAS disks. Of course you are right that 15k SAS disks are better than 7k NearlineSAS disks ( i think that IOPs are the reason), but only marginally. So my conclusion is that SATA disks has different electronics (that's the difference between NearLine SAS and SATA, NearLine SAS has the capacity of SATA but the SAS interface board) and there are some translations from SAS commands to SATA and that causes such speed degradation (this may happen in controller). But the NearLine SAS are real SAS drives so there's no such translation. 1 out of 1 found this valuable. Do you? |
