Anyone with HA and 5.7 ?

[Anatoly (staff)]

Not a problem! Please use this link to download.

[rchisholm]

We're running 5.7 with HA. Here is a 2 minute SQLIO random write with a 4GB file to an HA target with 2GB cache. This was performed within about 10 seconds of another identical test to simulate heavy jobs being run one after another.

Config:
Database Server: 2008R2 Enterprise on HP BL465c G7 with 1 X 12 Core Opteron 2.2GHz and 32GB RAM, 2 Flex Connect iSCSI HBA's set to 6Gb in MPIO RR.

StarWind Servers: 2008R2 Enterprise, 2 X 4 Core Xeon 2.4GHz and 72GB RAM, 2 X HP Dual 10GBe NC522SFP+ with 2 ports iSCSI and 2 ports teamed for sync

Targets: 1TB with 2GB write back cache. LSI 9280-8e RAID controller with 512MB battery backed cache with 40 Seagate 146GB SAS 15K drives in RAID60 and 8 Intel 32GB X25-E SSD's for CacheCade read cache.

C:\Program Files (x86)\SQLIO>sqlio -kW -b64 -frandom -s120 -LS -BN -o4 -Fparam.t
xt
sqlio v1.5.SG
using system counter for latency timings, 2148496 counts per second
parameter file used: param.txt
file d:\testfile.dat with 8 threads (0-7) using mask 0x0 (0)
8 threads writing for 120 secs to file d:\testfile.dat
using 64KB random IOs
enabling multiple I/Os per thread with 4 outstanding
buffering set to not use file nor disk caches (as is SQL Server)
using specified size: 4000 MB for file: d:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 4620.33
MBs/sec: 288.77
latency metrics:
Min_Latency(ms): 1
Avg_Latency(ms): 6
Max_Latency(ms): 219
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 0 1 13 23 20 12 7 4 3 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 4

[alexey (staff)]

Hi Anton,

Raw image without formatted, I can achieve 117MB per sec.

With formatted image, if I let IOmeter created a 4GB testing file on the drive, I would only get 3-6MB per sec.

What's the difference ? I can copy file into my iSCSI drive with blazing speed( bc of cache) but within the drive, if I would to copy stuffs out and make another copy of it on the same drive - transfer rate is only 3MB-6MB, which coincided with the same result of IOmeter.

Something went wrong ?

Hi Oxyi,
Can you pls tell me wich iometer settings were used to receive the result 3-6MB per sec (block size, % random/sequential, % read/write, number of outstanding IOs). May I ask you to perform similar tests with other benchmark utility? for example ATTO Disk Benchmark

[oxyi]

Hi Alexey,

64 I/Os, Siz is 8KB, 100% access, 65% read 60% random Burst 1

Here is the default setting for ATTO Disk.

[anton (staff)]

Performance is dead. Is there any chance one of our engineers could jump on your setup remotely to check what's wrong?

[oxyi]

sure, anytime

please have an engineer check my setup.

Case # 00004597: slowness StarWind Software Inc. ref:00D8LZUy.500CFlIet:ref

Thanks

[Anatoly (staff)]

I have contacted you regarding to this case. I`ll contact you in a minute via email.

[oxyi]

Just did some tweaking , I am able to stabilize IOmeter setting with 100% read , 100% sequential, 64 IO.

But if you see the network graph, the peak is at 115MB/sec, but it dropped off to 0.
Anyone be able to explain why ?

[mooseracing]

Dumb question but how are you guys testing your images on an MS Cluster box? I run tests against the Quorom disk as that is the only one with a drive letter assigned. I'm curious as to what everyone else is doing without taking their servers down?


edit: I'm running 5.5 right now and saving the I/O results from the post config file. Working on upgrading to 5.7 today hopefully.

[Anatoly (staff)]

I'm sorry, but could you clarify your question - what testing are you talking about? In case of testing failover features you can simply reboot one of the SAN servers or restart (or stop-start) StarWind service, if talking about testing purposes please refer to our Benchmarking guide.

[mooseracing]

I'm sorry, but could you clarify your question - what testing are you talking about?.

Running IO meter before you upgrade without coming in on a weekend, on a MS CLuster setup. Since on the cluster the drives become a link under %OS Drive%:\ClusterVolume . The only ways I can think of offhand are creating a VM on the cluster and testing it on that VM, or using the Quorum disk since it is presented as a drive. But I am relatively new to the cluster world so I may be missing something.

Also 5.7 went in smooth, I like seeing the performance tab. Hopefully more features like this in the future that can help us smaller places be able to log into the server and get a quick handle on what is going on.

[Anatoly (staff)]

Well, testing from StarWind side is simple here: put HA device to CSV (as generic, not quorum), stat installing VM and reboot one of the SAN node or restart starwind service on it. If Installing process is OK - your Highly Available storage is working properly.

[starneo]

I'm sorry, but could you clarify your question - what testing are you talking about?.

Running IO meter before you upgrade without coming in on a weekend, on a MS CLuster setup. Since on the cluster the drives become a link under %OS Drive%:\ClusterVolume . The only ways I can think of offhand are creating a VM on the cluster and testing it on that VM, or using the Quorum disk since it is presented as a drive. But I am relatively new to the cluster world so I may be missing something.

Also 5.7 went in smooth, I like seeing the performance tab. Hopefully more features like this in the future that can help us smaller places be able to log into the server and get a quick handle on what is going on.

You can "subst" the CSV path to a drive letter.
Use it like this: subst Y: C:\ClusterStorage\Volume1 - now you can benchmark this CSV

[starneo]

Now I have a question for my own.
We update from 5.5 to 5.7 last weekend.
Since that we encountered heavy load on the physikal raids on both Starwind Storages (it is a HA set), especially read I/O but write too.
We do have 4 targets each with 4GB Cache and one Target with 2GB Cache. The four targets are connected via iSCSI MPIO (4x1Gibt) as CSV to 4 (late 7) Hyper-V Nodes.
May it be that Starwind is not using the Cache in the right way? We are using the default Value of 5000ms.

[starneo]

Some Benchmarks from our HA Set: