iSCSI performance unacceptable

HP Microserver with 4 x Seagate 5900rpm drives set as two software raid mirrors, both MDx in a single Volume Group, one big iSCSI Target, Intel e1000 PCI-e card, dedicated to iSCSI network

Attached to Win Server 2003 R2 via (also) dedicated iSCSI Intel NIC

The iSCSI speed is awful (write ~ 25Mb/s)

It is not Win Server 2003 fault, as is also connect via the same iSCSI network to Thecus i5500 unit

OK, the Thecus has 7200rpm drives & build in HW RAID

But the hardware is now 3 years old, yet it blasts write of excess 100 Mb/sec (that is at least FOUR times faster)

What to do to improve the HP/DSS combo?

Thanks

sebus

Do you have the latest build of DSSv6? as you may need a new driver for your RAID, so update to the latest build could help you to get the latest drivers needed.

Also try to tune your system, you can find that in the KB at:
http://kb.open-e.com/42/
and
http://kb.open-e.com/How-can-we-impr...eter_1083.html

Sebus,

I am not surprised that your performance is poor.

The miniserver has just about the worse posible setup.

You are using slow HDDs. The Seagate 5900 RPM drives have read and write seek times of 12 and 13ms, compared to 8.5 and 9.5ms for 7200 RPM drives, or 3.8/4.4 ms!! for 10,000 PRM drives.

You are using *Software* RAID

Further,

The HP Microserver is running a generic AMD Athlon II CPU and generic Server NIC.

Whereas, the Thecus devices is a *specialized IO processor* (Intel® IOP331 I/O) which has integrated RAID 6 and iSCSI CRC32C functionality.

A specialized device will always outperform a generic one.

NO, the HP does not run a generic NIC, as I replaced it with PCI-e Intel version, so I can get jumbo frames (the on-board Broadcom does not have jumbo frames)

And what is wrong with software RAID?

I am sure that specialized device will outperform, but 25Mb/s I can get on USB2

On same hardware HP Microserver (with same drives, also softrware RAID, and even on-board Broadcom, as I did not yet swap it) with Openfiler 2.3 I can get FTP writes of excess 60 Mb/s

And yes, I have the latest available version of DSS 6

I do not expect miracle (like my Equallogic array), but twice the current speed would be more like what should be expected

sebus

Quote:

---

Originally Posted by Al-S

Also try to tune your system, you can find that in the KB at:
http://kb.open-e.com/42/

---

I do NOT have AutoNeg parameter on that screen to edit

sebus

Try this:

1. From the console, press CTRL+ALT+W,
2. Select Tuning options -> iSCSI daemon options -> Target options,
3. Select the target in question,
4. change the MaxRecvDataSegmentLength and MaxXmitDataSegmentLength values to the maximal required data size

(check w/ the initiators to match).

maxRecvDataSegmentLen=262144
MaxBurstLength=16776192
Maxxmitdatasegment=262144
FirstBurstLength=65536
DataDigest=None
maxoutstandingr2t=8
InitialR2T=No
ImmediateData=Yes
headerDigest=None
Wthreads=8

These settings do make the speed jump to a more acceptable 42Mb/s on write

sebus

At this point, without hardware raid, your at about the top for tuning your setup.

With hardware raid, there are more things to tune for speeds.

One other thing to use is block I/O and WT on the targets, if not already.

For a D2D2T backup it will have to do, no space to fit RAID card

Why these options are not default then?

sebus

Since the daemons that run the connections are open source, and we don't know with certainty what each system may involve, or hardware used, we stay on the defaults to start with.
But for backups, that's not to bad really.

Make sure the LUN has the Write Back enabled. Also check the test.log and tell me what the hdparm are for the reads - cut an past on your post.

Also SeanLeyne is correct that those drives are slow "Seagate 5900 RPM" with the seek times at that rate dont expect great performance values especially again a RAID controller with faster drives again slow drives with no RAID controller to tune with the Write Cache that is available with the RAID controller.

Just for the fun I build another unit with exactly the same spec & this time used FreeNAS FreeNAS-8.0-RELEASE-amd64.Full_Install

Default install with same tweaks as per Gr-R recommendation

Both servers connected to the same switch via identical Cat6 cable

So all in all the hardware is a TWIN, just the OS is different

Then I used Copy Handler 1.32 to copy same 1.59Gb of data from internal HD to both units:

First FreeNAS:

45 sec, avg 36.26 MB/s

Then DSS:

4 min 16 sec, 6.37 MB/s

Surely not what one should/could expect?


And that is exactly what I see during normal use of the machine.
Open-e ISCSI volume can not be accessed when something is writing to it

With FreeNAS I can read (open large rar) while another process writes to the volume (ie file copy)

sebus

Can you check the test.log and tell me what the hdparm are for the reads - cut an past on your post.

And where is test.log?

sebus

Status > Hardware > Log then click on the Download button and open the log file and look for the test.log

Or learn here:

http://www.open-e.com/library/webcasts-and-videos/

2010.11 How to view and understand the log file from DSS V6

Quote:

---

Originally Posted by Gr-R

Try this:

1. From the console, press CTRL+ALT+W,
2. Select Tuning options -> iSCSI daemon options -> Target options,
3. Select the target in question,
4. change the MaxRecvDataSegmentLength and MaxXmitDataSegmentLength values to the maximal required data size

(check w/ the initiators to match).

maxRecvDataSegmentLen=262144
MaxBurstLength=16776192
Maxxmitdatasegment=262144
FirstBurstLength=65536
DataDigest=None
maxoutstandingr2t=8
InitialR2T=No
ImmediateData=Yes
headerDigest=None
Wthreads=8

---

As an update, please keep in mind when using latest builds of DSS, MaxBurstLength variable cannot be set to 16776192 because such value was only supported in iSCSI Enterprise Target (IETD), additionally With reads variable also works only with (IETD).

Max value in SCST for MaxBurstLength is 1MB, 10485760.

In case when you use MaxBurstLength=16776192 and make the update, scripts during starting the system fix the values to default one (262144).

So you may leave MaxBurstLength variable values as the default value or you may use the max value which is 10485760.

@Al-S

You got a bad Typo there:

Quote:

---

Max value in SCST for MaxBurstLength is 1MB, 10485760
.
.
So you may leave MaxBurstLength variable values as the default value or you may use the max value which is 10485760.

---

should read:

Max value in SCST for MaxBurstLength is 1MB, 1048576
So you may leave MaxBurstLength variable values as the default value or you may use the max value which is 1048576.


Just one "0" to much...

Regards,
Ralph