StoreEver Tape Storage
cancel
Showing results for 
Search instead for 
Did you mean: 

HP MSL 2024 SCSI 1840 drive throughput

SOLVED
Go to solution
KEN_SYD
Frequent Advisor

HP MSL 2024 SCSI 1840 drive throughput

Hi All,

I am using MSL2024 SCSI 1840 drive to backup 6-7 TB data from SAN every friday. My everage backup throughput is 6.2GB/min to 6.4GB/min via 4 Gigabit NICs using arcserver backup software with multiplexing option.

Hardware commpression is enabled. I notice that each LT04 only hold around 1.2 TB, that means I am archiving compression ratio 1.5:1 with my current data.

Based on MSL2024 specification, data transfer rate for commpression 2:1 is 864 GB/hour. Which mean I should archive max 648 GB/hour or 10.8 GB/min for compression 1.5:1. am I correct?

Is there anyone here that archive better throughput than me so I can do a comparison?
I and my team tried to find out why we are running at max 6.4 GB/min from SAN with 4 NICs.

We are not going for backup from SAN via FC using VDDK as all vmdk files will be backed up and that is a big amount of data.

Thank you.
39 REPLIES
Marino Meloni_1
Honored Contributor
Solution

Re: HP MSL 2024 SCSI 1840 drive throughput

if I well remember, the lto 4 have a transfer rate of 120 MB/s native, this should be the max he can write to the tape itself, and the speed is limited by the tecnology and the motor.

you are seeing 6.2 GB/min that is 103 MB/s


now seeing these number that looks apparently good, 103 vs 120, but in fact I do not know where and how you are checking the troughput

apparently if you reack 6,2 GB/min = 103 MB/s with a Compression ration of 1.5 you should be able to reach a troughput of 103*1.5 = 154 MB/s = 9.2 GB/s

I have not very clear how the library is connected, but in order to test if the reduction is comming from the data source of from the target (the msl2024) you can use LTT with the "dev perf" test.
This test will generate data in the host memory and via the scsi bus (or FC bus) will send the data directly to the Drive, showing the effective speed you can reach in your configuration (host, HBA, scsi of fc path and Drive)
I suggest to use a amount of data that run for at least 15 minuts in order to avoid false report due to buffer etc...
If the result reach the expected value (120 MB/s with 1:1 compression or 240 MB/s with 2:1 compression) then the problem is with the data source, or data source path.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Thank you so much for your response with interesting information, Marino.

Could you please give me more details on LTT with "dev perf" test?

9.2 GB/min (not GB/s) is what i'm targeting for.
My current tape lib connected to backup server via SCSI interface.
The currently average throughput is reported from ArcServer backup program. It reports how much data has been backed up, how many files and throughput.

Thanks again.
Jens Ey
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Ken,

personally, I find this throughput rate very good. I am stuck with 2.5 GB/min from local disk arrays.

If you try to match your total throughput rate with your compression ratio you should have in mind that both are average values while 864 GByte/h is a absolute maximum.

First you loose a bit due to soft write errors, second it might be (and that ist very likely) that the compression ratio of your data is not constant. So it might be that sometimes the date from SAN are coming too fast to write them to tape w/o loss of throuput and sometimes the tape could write more data if even the SAN would deliver it in time.

Hope that helps a bit,

Jerns
Marino Meloni_1
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

when you start LTT, and you select one drive, in the upper area of LTT windows, you will see one button that says "dev Perf"
this is what you need to select. In the left panel, then you will have the possibility to select options, here you can choise some parameters, one is the time or the amount of data you want to use for the test.
The test will overwrite your tape, so use a scratch one, if possible a brand new one, or a know good one, as a bad tape can inpact your performance too.

Yes, typo, 154 MB/s = 9.2 GB/M

related to the mesurement, if you got it from ArcServer, then this is mesuring the data source, and it is uncompressed data.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Thank you everyone.

Thank you for introduce me LTT software. I tried HP Tape Assure but it could not detect my drive with lastest firmware.

I will try with LTT and report the result soon.
CLEB
Valued Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

Ken

I have a DL580 G5 connected to an MSL2024 1840 SCSI via a SC11Xe HBA.

Using ARCserve R15 to backup some 10GB files from a local RAID5 partition yields me an average throughput of 5.174GB/min.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi CLEB,

5.2 GB/min from local backup is good throughput I think.

In my case, I had to use 4 NICs and backup multiplexing from 4 sources simultaneously to get average of 6.2 GB/min.

Also, throughput is higher when backing up somes big files compared to thousand small files.

Will update after doing testing with LTT.
Curtis Ballard
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

An often overlook tuning parameter is the transfer size your application is using. To hit the maximum transfer rates you have to use at least a 256K transfer size.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Yes, I'm testing with different size at the moment.
I tried from 48, 96, 128, 512 already. unfortunately, average throughput is almost the same. This week full backup will be run with 256.

Arcserve explained that, small size mean software tries to send packet immediately to keep data stream constantly rather up and down.

Let see how it's going with 256, but I guest throughput will remain the same.

Thank you for you suggestion.
CLEB
Valued Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

Ken

I did some more testing and managed to get 6.7GB/min maximum.

Are you using arcserve too? I have the default transfer size which is 64K. Is this configurable?
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi CLEB,

I'm currently using Arcserve R15.
This value can be changed depend on each environment. You may need to call up CA support for help to change this value because some other values may change along with MaxSendPacketSize.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi All,

I've done couple of test with LTT 4.11 Dev perf with latest driver & firmware for LTO4 drive.

With Block mode Fixed ans test size 16 GB. Speed is 177 MB/s for all 2:1, Zero and 1.6:1 compression. (Note: hardware compression is on)

Why I got the same speed for diferent compression? It should be around 120 MB/s for Zero compression and around 240 MB/s for 2:1?

The LTO4 tape that I used for the test is brand new HP tape.

Could someone give me some sugessions what to test more to confirm what is the real maximum speed that my drive can handle.

Thank you so much.
Marino Meloni_1
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

hi
in the directory log in the storageworksltt path, there should be a file called resultlog, can you attach it here?
the test allow you to test the path between the host and the drive,

if your current througput is limited to 120 MB/s indipendently of what is the compression ratio, that mean this is the max throughput on the path.

the drive can handle 120 MB/s with 1:1 compression ration, and 240 MB/s with 2:1 comp ratio, this because the decompression is made on a chip onboard, before to write phisically on the tape (the max troughput writing on the tape is 120 MB/s due to the motor and tecnology, and this cannot change)

during the test, did you selected the whole tape?
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Marino,
Thank you for your response.

Attached file is resultlog.ltt

I did select test size is 16 GB, and speed was 177 MB/s for 2:1, 1:1 and 1.6:1 compression.

Do you reckon I need to select full tape for best speed test?

Looking forward to hear from you soon.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Marino,

Is there anything interesting from the log?
I disabled hardware compression and did a test again.
This time I got 110 MB/s for all 2:1, 1.6:1 or zero compression. Throughput pretty much the same for test size 16GB, 48GB, 64GB.
Block size and I/O size seems no effect.

Waiting for your comments.
Marino Meloni_1
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Ken

I thought I had answered to the question, but something went wrong and my replay is not attached, sorry

From the log I can see that some tapes are marginal, during previous test, the margin in R/W mode was lower then 100%, this is another factor that may influence the throughput

Also I was suggesting to run a test with at least 15 minutes of I/0 in order to get rif of any buffer effect, the tests you runs where 1.30 minutes each

177 MB/s looks to me the max you can acchieve on your SCSI path, from the host to the drive, only way to improve is to try tuning the block size, and also be sure the scsi HBA do not have shared resources with other server components.
You may try to reset the server config to default, or change the slot of SCSI HBA it may be secondary bridged PCI bus (long time I didn't work with server hw, may be nowadays these kind of issue do not exist anymore)

anyway, 177 MBS for the SCSI subsystem is not so bad as it will be hard you can reach the same throughput from the datasource

If you want LTT have also a way to test the data source troughput, the test is "Sys Perf", you can run a Backup performance test, directly on the data you have,
you also have the possibility to create your own data, using the restore performance test, this will allow you to have data you know location and quality/
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

I greatly appriciate your quick response Marino.

I will do more test next week. By the way, my SCSI is ultra320 (320 MB/s) so max throughput on SCSI is much higher than max drive throughput (240 MB/s compress 2:1).

Also I found that the compression ratio in Dev Test option is no meaning with my test. I got 177 with drive compression and 110 without drive compression regardless ratio 2:1, 1.6:1 or zero. Is that correct?

Have a nice wk.
Marino Meloni_1
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

Several factors may limit the troughput, that may be:
-the server itself: lack of resources, or other task running inbackground
- the phisical layer: the scsi cable, the scsi terminator, the drive scsi interface, any of the components on the scsi chain may be limiting the whole communication.
- traffic congestion: polling for example comming from the host, or other supervisor running on the host (insight agents, or similar) these create traffic on the scsi bus, and the drive should answer to these scsi command causing delay and overhead


you may see a variation of the troughput during test changin the compression ratio when nothing is limiting it, in your case something (as just explained) is limiting.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Marino,

I'm back after opening a case with HP.
After checking my support ticket, they found that polling TUR is 10 per second which to much and send me a document how to disable TUR polling.

I applied registy AutoRun = 0 for hplto key, but don't know how to check the TUR. Email back to HP support but not reply yet. I think you may response quicker.

Which tool can I check TUR? After disable autorun, test throughput remains the same.
Thanks.
Marino Meloni_1
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

you can check the polling collecting a support ticket.
you can check the polling setting details to "everything"
then select "Classic support ticket"
and "enhanced SCSI COmmand History"

here you can see the scsi commands and the time of the events

Marino
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Thank you for quick response.

However, can you instruct me where to find these settings and classic ticket? From LTT or Tape Library?

Then I need to send the ticket to HP engineer right?
Curtis Ballard
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

I can't explain the strange behavior with reporting the same speed for different compression ratios with your drive however if you are able to get 177MB/s out of a parallel SCSI system then it is extremely unlikely that anything at the drive end is causing your performance problem. 177MB/s over parallel SCSI is quite good and most systems I have worked on can't hit that speed regardless of what type storage device is on the other end of the bus.

Disabling the polling is good but unlikely to help much as the same polling would have been going on when you ran the performance test.

To check for the polling you need to use L&TT, select the "support ticket" button, select "Extract device data", then select "view" then follow the instructions you were given earlier to find the data in the ticket.
KEN_SYD
Frequent Advisor

Re: HP MSL 2024 SCSI 1840 drive throughput

Hi Curtis,

Thank you for you comments.
Is speed with or without compression the same because of hardware compression is enable?
Curtis Ballard
Honored Contributor

Re: HP MSL 2024 SCSI 1840 drive throughput

I reviewed the log file you attached with your performance testing. I see results for 2:1 compression at 177MB/s, results for all zeros at 177MB/s, and results for 1.6:1 at 177MB/s.

All of those make sense. The drive is capable of 192MB/s with 1.6:1 compression so if the system tops out at 177MB/s then 1.6:1 and higher compression ratios will all yield the same results.

I also see results for uncompressed data at 111MB/s which seems reasonable with the drive maximum speed when everything is perfect being 120MB/s.

The results in that test look reasonable to me. I did see that there were conflicting results from the drive assessment test. I would say that the tape used for the first test probably wasn't quite perfect. Since the second test passed as long as the drive is working correctly I wouldn't worry about the first test issue.