Back up times. How long 1

[Teknoratti]

How long does it usually take for your weekend backups to complete? Of course this is barring any mishaps.

I need to find out why it is taking an enormous time for our backups to complete. Our backups usually start around 7:00pm on Friday, and come Monday they are barely 50% complete. I think I will decrease the amount that is backed up on the weekend, and or figure out how to increase the throughput.

It seems very strange. On one of our servers the average throughput is only 39 MB/MIN, on a server that has over 45
GB to be backed up (45,229.50 MB). The elapsed time is about 19 hours, 20 mins.

The throughput on the other servers seem about normal.

 

[vschumpy]

How long is a piece of string ?

Sorry, but figures mean nothing unless you put some specific details of your environment against them ie hardware, what sort of data you are backing up etc and how.

 

[NickFerrar]

Yeah it varies dramatically depending on what you're backing u and what you're backing up onto. Lots of small files (user profile areas etc.) take a lot longer to backup for the equivalent size of big files.

There are some registry you can alter to improve performance though, I think regarding the block size. Not done it myself though. We typically backup 45GB in around 30mins-1.5 hours depending on if it's to DLT or LTO (and what the source files are). We have one server with an LTO-3 drive that backs up 140GB in 1hr 12mins.

 

[vschumpy]

There are no ARCServe registry tweaks for speed. You should never need to touch blocksize ever in ARCserve these days. If you do your device was probably never certified with the version of ARCserve you are using.

 

[bobarc]

Turn on multiplexing.

 

[vschumpy]

Whether multiplexing has any effect would depend on the following:

1 Whether the version of ARCserve being used offers that function
2 The backup is remote - multiplexing is absolutely useless for local backups.
3 Whether the hardware has any spare capacity to benefit from the feature in the first place.

It all comes down to needing more information to be able to help with this query.

 

[cokeonice]

Can I hijack this thread?
The original posters scenario appears to be the same as mine.

Bought a Dell 132T LTO3 to replace Dell 122T LTO1 because we wanted to cut our backup times by 1/4th. According to base specifications, that should happen without anything else changing.

What do I need to add?

All servers (except one remote one on a different subnet which is a domain controller) are plugged into a Gigabit switch and have Gigabit ethernet cards.

The server that runs Arcserve is a Dell PE2650, 2-2.8GHz Xeon, 2Gb RAM, 1Gbps ethernet, Adaptec 39160 Ultra160 SCSI card. Tape drive is on own channel.

This server is also an application server.

There are a total of 17 servers being backed up by Arcserve in one job.

Server Description
2 Win2000 Server Advanced. SQL Server, PE6600, 4-P4 2.0GHz, 8Gb DDR, 2- 1Gbps Ethernet connection (bridged as one). This server has 147Gb of data to backup of which 121Gb is related to SQL.

3 Windows Powered? Storage (NAS) device. Dell Powervault. P4 2.6GHz, 512MB RAM, 682Gb total storage space of which 171Gb is being backup up (Image files).

4 Win2000 Server Standard. Webserver. This thing isn't even a server. Dell. P4 1.7GHz, 1Gb RAM. Not much data being backed up.

5 Win2003 Enterprise Server. SQL Server, Dell PE6600, 4-2.0GHz, 8Gb DDR, 1Gbps ethernet connection. This server is a test SQL server with roughly 80Gb of data.

6 Win2003 Server Standard. SQL Server, Dell PE2800, 2-3.6GHz Xeon, 4Gb RAM, 1Gbps ethernet. This server has very little data to be backed up. New server. Mainly the C: drive.

7 Win2003 Standard Server. Communications server (runs a telecom product for customer service). Dell PE2800, 2-3.0GHz Xeon, 2Gb RAM, 1Gbps ethernet. Not much data backup up from this server. Mostly the C: drive.

8 Win2000 Server Standard. Domain Controller, fileserver. Dell PE2600, 2-2.4GHz Xeon, 2Gb RAM, 1Gbps ethernet. There is approximately 180Gb of data being backed up from this server.

9 Win2000 Server Standard. SQL Server. PE2600, 2-3.0GHz Xeon, 4Gb RAM, 1GBps ethernet. This server has roughly 150Gb of data to backup.

10 Win2003 Server Standard. Will be new webserver. PE2600, 2-3.0GHz Xeon, 2GB RAM, 1Gbps ethernet. No data. Just system being backed up at this point.

11 Win2000 Server Standard. SQL Server. PE2600, 2-3.0GHz Xeon, 4Gb RAM, 1Gbps ethernet. This server has approx. 50Gb of data and system to backup.

12 Win2003 Server Standard. No job yet (will be new mail server). PE2600, 2-2.4GHz Xeon, 2Gb RAM, 1Gbps ethernet.

13 Win2000 Server Standard. Image fileserver. PE2600, 2-2.4GHz Xeon, 2Gb RAM, 1Gbps ethernet. This server has 160Gb of data and system files to backup.

14 Win2000 Server Standard. Exchange server. PE2600, 2-1.8GHz Xeon, 2Gb RAM, 1Gbps ethernet. Mail server with a total of 60Gb of data and system files to backup.

15 Windows Powered? NAS Device. Dell Powervault. 2.6GHz P4, 2Gb RAM, 1Gbps ethernet. This NAS device has approximately 200Gb of data and system files to backup.

16 Win2000 Server Standard. SQL Server. Gateway 8400 Server, 4-PIII 700Mhz, 8Gb RAM, 1Gbps ethernet. This server has approx 90Gb of data and system files to backup.

17 Win2000 Server Standard. Fileserver. Gateway 7400 Server, 2-PIII 800Mhz, 1Gb RAM, 1Gbps ethernet. This server has roughly 80Gb of data (images) and system files to backup.

The above servers are on the same subnet, plugged into the same Gigabit switch.

18 Win2000 Server Standard. Domain controller/fileserver. PE2600, 2-1.7GHz Xeon, 2Gb RAM, 1Gbps ethernet. This server has only 25Gb of data and system files to backup. This server is on a different subnet than the rest.

My situation is similar to the other guys...backups start at 8:00PM on Friday, end around 11:00AM on Monday.

 

[davidmichel]

depends on were the bottleneck is. If it is the network then using tape drives with greater throughput will not improve things.

Start by checking the throughput for each session to see the average, the high, and lows.

 

[vschumpy]

Interesting comment "Bought a Dell 132T LTO3 to replace Dell 122T LTO1 because we wanted to cut our backup times by 1/4th. According to base specifications, that should happen without anything else changing."

This is not necessarily true. Information why upgrading to LTO3 may not increase, and may even decrease backup throughput:

http://www.techworld.com/storage/features/index.cfm?featureid=1125&Page=2&pagePos=5

 

[vschumpy]

.... Also if you are trying to MUX all 17 servers in one job, the overhead will be way higher than any performance gain you might potentially achieve IMO. Try halfing the number of concurrent streams to 8 and see if the performance improves.

Also potentially the PIII machines could be a bottleneck - try dropping them out into a separate job that runs after the main one (yes, even with the amount of memory they have).

 

[cokeonice]

vschumpy,

Good article. It confirms some of the other statements about slower systems possibly being part of the problem. Have to glean what you can from an article that begins to promote someones software.

I don't understand the MUX thing...well, I guess I might. Are you saying, ArcServe is attempting to pull data from all servers simultaneously? From the job log, it appears that it is only reading from one server at a time.

This type of backup technology is new to me so please bear with me.

CokeOnIce

 

[vschumpy]

If it's only pulling from one server at a time you are not using multiplexing.

If you are backing up servers remotely across the LAN it's possible your not able to stream sufficient data to keep the drive running without playing catchup as described in the article.

If you have ARCserve 11.x then you could try using multiplexing if (as you sort of hint at) you are not already (You can run MUX with 4 concurrent streams as standard in the base product without having to purchase any additional licences).

A small whitepaper on Multiplexing here should help you with the basics:

http://www.ca.com/channel/emea/bab11_1/docs/wp_multiplexing_brightstor.pdf

 

[cokeonice]

That article was pretty good. But it looks like I can only do four servers per job?

I will have to look into that.

If I have to break the servers out into jobs of four, I will have to figure out how to run them concurrently?

 

[vschumpy]

No need to split them out, it will run 4 concurrent streams then when one has finished will add the next one until all the nodes are backed up. It doesn't just mean you are limited to 4 nodes per job.

If you really want to have more than 4 concurrent streams you can purchase and install the enterprise module which gives support for up to 32 concurrent streams, although I've never seen anyone use anywhere near that number of streams in real-life.

 

[cokeonice]

Thanks again vschumpy!

I have learned a lot from you!