View speeds for reading and writing to LTO

mosiac

ADSM.ORG Member
Joined
Sep 25, 2014
Messages
152
Reaction score
0
Points
0
PREDATAR Control23

I was wondering what everyone uses to view the traffic IO of data written to tape? I'm running a liunx TSM server and qlogic fc cards to a cisco san switch to the tape drives.

I can look at the san switch when all 4 drives are in use and it only shows on average 10 to 20 percent utilization and I feel like there's a major bottleneck I'm missing.

I believe my LTO5 drives are 8gb and I know my FC switch and FC card are both 4gb so I feel like I should be able to use a lot more bandwidth during operations.

Thanks,
 
PREDATAR Control23

In AIX I use iostat -p

Linux should have a similar command.

The switch utilisation isn't all that valuable here as the limitation is more likely to be the LTO5 hardware. LTO5 max speed is 140MB/s uncompressed. You can use SAN tools to measure speed but using OS tools should be just as good.
 
PREDATAR Control23

I'm checking this out but I'm not sure I'm seeing my fc card showing up it looks like just storage mount points.
 
PREDATAR Control23

Since I'm using Qlogic cards I thought I would give the QConvergeConsole a try. It seems to be giving me information based off what I have hooked up which is good but even though I'm seeing all 4 drives in use in TSM
"q mount f=d"
I'm only seeing around 10 percent utilization of the link they are all attached to.

I doubt my disks are my bottleneck but I think perhaps ram is at this point.

Thanks for the help.
 
PREDATAR Control23

Since I'm using Qlogic cards I thought I would give the QConvergeConsole a try. It seems to be giving me information based off what I have hooked up which is good but even though I'm seeing all 4 drives in use in TSM
"q mount f=d"
I'm only seeing around 10 percent utilization of the link they are all attached to.

I doubt my disks are my bottleneck but I think perhaps ram is at this point.

Thanks for the help.

What platform, memory and # of CPU are you running TSM on? What TSM version?
 
PREDATAR Control23

Running on RHEL 6, 68gb of ram and 2 Intel Xeon E5-2500 procs running at 2.40gz per core. TSM is 7.1.5.
 
PREDATAR Control23

RAM is not an issue as you have lots of it. You have 2 Fiber HBA and are each zoned to particular tape drives, or are these running in load balancing manner?
 
PREDATAR Control23

Well I have 2 fiber HBA's but I'm only using 1 port on one card because there's so little traffic going through. The ports are zoned to all tape drives.
 
PREDATAR Control23

Would it be better if I had a direct line from an hba to a drive I want to use. I only have 4 drives in use and I could adjust my zoning to be more 1 to 1
 
PREDATAR Control23

Would it be better if I had a direct line from an hba to a drive I want to use. I only have 4 drives in use and I could adjust my zoning to be more 1 to 1

If you have 4 ports, then a direct connect would be an option.

However, don't expect to get close to 4 GB (between 2 and 3 GB/s at best) out per port consistently as the other limiting factor is the PCI bus
 
PREDATAR Control23

I shouldn't have too much trouble since my PCI bus should be 10Gb. I am noticing that I'm getting a lot of Device Errors if I"m looking at the Qlogic Monitoring for the port I've been using the most. I have no idea how to troubleshoot that either.
 
PREDATAR Control23

Errors can be one of two things:

- firmware version
- hardware issues like fiber cable type, physical fiber issues, etc
 
PREDATAR Control23

Firmware is at the latest available for these cards. I can double check all the cables though thanks!
 
PREDATAR Control23

So I've checked my cables and everything seems to be the same after switching them out and restarting the Tivoli Service. I have noticed that after attaching all available HBA ports more "IBMTape" connections show up in linux but as far as Tivoli is concerned it's still only sending data through one link, and that maybe a reboot required type situation.

I still don't understand why 1 port on a 4gb card isn't getting more saturated sending data to 4 tape drives at once.
 
PREDATAR Control23

So all ports seem to be doing Point to Point connection according to the Qlogic Gui. Traffic has moved to one of the other ports I connected after reboot but it's still not spreading the load not that it matters because even with all 4 drives working a 4 different node session data (I don't understand why sessions are talking directly to tape drives) I'm still seeing almost no utilization.
 
PREDATAR Control23

So the flow of data is going something like below.

Client disk -> Client OS ->Client NIC->Network->TSM NIC->TSM OS->TSM HBA->SAN Switch->Tape Drive

You are looking at the SAN Switch and TSM HBA and thinking the utilisation is looking low. All the parts of the data flow are important a bottleneck may exist in any one of them or even multiple. Without knowing how fast the data is being written to tape, we don't even know if there is a problem. (yes, I would expect that one 4Gb HBA cannot drive 4 LTO5 tape drives full speed)

Personally, I'd start with iostat to measure how fast data is being sent to the tape drives. If it seems slow still then see how fast it is in a tape to tape operation. Results from this will give some idea on where to look.

Just looked at Linux options for tape drive stats. This looks helpful - http://prefetch.net/blog/index.php/...ut-and-performance-statistics-on-linux-hosts/
 
PREDATAR Control23

So the flow of data is going something like below.

Client disk -> Client OS ->Client NIC->Network->TSM NIC->TSM OS->TSM HBA->SAN Switch->Tape Drive

You are looking at the SAN Switch and TSM HBA and thinking the utilisation is looking low. All the parts of the data flow are important a bottleneck may exist in any one of them or even multiple. Without knowing how fast the data is being written to tape, we don't even know if there is a problem. (yes, I would expect that one 4Gb HBA cannot drive 4 LTO5 tape drives full speed)

Personally, I'd start with iostat to measure how fast data is being sent to the tape drives. If it seems slow still then see how fast it is in a tape to tape operation. Results from this will give some idea on where to look.

Just looked at Linux options for tape drive stats. This looks helpful - http://prefetch.net/blog/index.php/...ut-and-performance-statistics-on-linux-hosts/


I agree with everything you say but the path connectivity looks like this
Client Disk (10gb netapp) -> client os -> client nic (10gb vnic) ->Fabric Interconnect (40gb links between chassis') ->TSM Nic (10gb vnic) -> TSM HBA (4gb FC card) -> San Switch (4gb) -> Tape drive 4gb.

It looks like Tivoli is currently only using 1 port to send all traffic to the tape library, the Qlogic dash is saying the effective speed of any one port on my hba is actually 2.5Gb. This morning I attempted to zone a single LTO5 drive to a single HBA port so that 2.5 could be used specifically for those processes and the zoning didn't work. I'm pretty sure I messed something up, but I'm wondering about MPIO settings I don't see any listed in my TSM documentation but that doesn't mean my documentation is right at all.


Thanks for the help so far everyone.
 
Top