Unsolved
This post is more than 5 years old
30 Posts
0
856
analysis of speed of backups (ndmp, client backups using advfile)
speed
24 savesets for NDMP (3 different filesystems from Celerra host - 8 sets used each). Averaged 11MB/s backing up to Adv File Devices. Each saveset averaged 304GB in size
90 savesets for server back-ups: Averaged 2.32MB/s backing up to adv file. Each saveset averaged 7GB in size
ble1
2 Intern
2 Intern
•
14.3K Posts
0
September 28th, 2006 13:00
Celerra host - 8 sets used each). Averaged 11MB/s
backing up to Adv File Devices. Each saveset
averaged 304GB in size
From above I assume you are using network backup. With move to new tape technology I would suggest to assign tape device to data mover. We recently tested Celerra via VBB (block size backups as opposed to traditional file structure ones like tar/dump aka PAX) and results were very good. We used LTO3 and were able to get 60-80(max) MB/s. With tar/dump we would get 10-20MB/s lower values. Having 11MB/s average rings the bell of 100MBit limit of NIC or switch port so unless you have greater values then perhaps there is space to improve those stats.
2k3 servers:
90 savesets for server back-ups: Averaged
2.32MB/s backing up to adv file. Each saveset
averaged 7GB in size.
Now that's poor, but it could be that either box can't push data or something within network configuration is not configured correctly. To get to the bottom of that you would need to run several tests that would quickly show you where your bottleneck is (disk speed, network, CPU, networker configuration, etc..).
back-ups. I'm wondering if the time that's marked
for completion is when the entire GROUP finishes.
(is that correct?) - regardless of this though, the
90 savesets used in this sample were from the same
group, so when averaged, perhaps still yield
accurate average speed. (although I didn't include
the ASR, Index, related to the group)
90 savesets might be nothing for one group, but it could be too much as well. It really depends on how good your storage node is sized and how does it cope with load.
backing up to disk...I'd think be faster.
Agreed - you should see much faster values. I tend to believe you have bottleneck there due to some physical limit or bad configuration.
stating "160MB/s non-enrypted performance". (80MB/s
w/ encryption) -- does this sound like a "marketing
speed under complete ideal conditions".
Sounds like LTO4 with speed of LTO3 and yes that would be markering speed. Right now I still see LTO2 better than LTO3 when it comes to calculate the cost of how much is required to invest to push those drives. If you can't push LTO3 with at least 35MB/s don't even consider such media type. Now, LTO3 has same max specs - 80/160, but to drive that to let's say 120MB/s you need:
- fast bus
- CPU power
- at least 2gbit HBA
... and that would for just one drive. Bigger is better, but than there is always another side of the coin.
bottleneck seems to be my NDMP source? (or is my
current bottleneck my AdvFile Devices?) ALso, at
160MB/s - woudnt my SCSI bus be saturated?
You'll have to run tests to see where your bottleneck is. Both file system backup and NDMP backup are giving you poor speed. What is the speed you get from staging?
ble1
2 Intern
2 Intern
•
14.3K Posts
0
September 28th, 2006 13:00
back-ups. I'm wondering if the time that's marked
for completion is when the entire GROUP finishes.
(is that correct?)
Yes if you are looking at the savegroup completion report. Look at mdb data via mminfo for each saveset's size, start and end time. You may find that some savesets are quick and some are rather slow.
davise
30 Posts
0
September 29th, 2006 09:00
This is a good question and was my next investigation. What is the best way to get this value? Look at the daemon log? or use a stopwatch? Look at the device speed in the GUI?
I got to thinking about what you said regarding network saturation. If I am in fact on 100Mbs network, will my back-ups never go beyond 11MB/s? (I'm slow at math, so I I thought to ask) I'm double checking the NICs, routers, and switches with my boss, (since he handles that), but I thought we were on GiGabit. (and Windows tells me i'm at a Gigabit on the networker server, but might not be on the Celerra end)
This network cap got me thinking about the poor performance on my Server backups. So, if that's true, my 2.3MB/s on my servers savesets may make more sense since i have a networker parralism of 4 (and when that group runs, they do so in 4's). So, 4 x 2.3 = 9.28 , which although is not 11, it's close enough. (and those numbers are averaged too)
Thoughts?
ble1
2 Intern
2 Intern
•
14.3K Posts
0
September 30th, 2006 14:00
investigation. What is the best way to get this
value? Look at the daemon log? or use a stopwatch?
Look at the device speed in the GUI?
Either of 3 will do.
network saturation. If I am in fact on 100Mbs
network, will my back-ups never go beyond 11MB/s?
They won't.
davise
30 Posts
0
October 2nd, 2006 08:00
However, with that being said, when i do a simple Windows file copy of a 1.6GB file. From both server to server, celerra to server, both directions I get a slightly faster speed, but still 12-15MB /s.
Do you still think there is somethign goofy going on (speed wise), with straight windows copies of 12-15MB/s?
ble1
2 Intern
2 Intern
•
14.3K Posts
0
October 2nd, 2006 09:00
(speed wise), with straight windows copies of
12-15MB/s?
Write down on paper path data from one server does to another and then identify components on that path and test each of them. That usually means testing disk read performance first (on the client).