theboyk Posted July 3, 2012 Report Share Posted July 3, 2012 I've been building a new Retrospect server for our studio and before deploying the new system, since I had the resources available, I ran some speed tests to see how Retrospect 9 performed on the new server within our environment (we're currently running Retrospect 6 and this new server will be running v9). Based on my testing (details below), when I enable software compression, it results in the backup process taking twice as long as it takes when there is no compression, or when hardware compression is enabled. Is this normal? Should software compression take twice as long as no compression or hardware compression? Details on the testing (which took place over the period of roughly 9 hours — fun day!) are as follow: Retrospect Server Machine Details: - Intel Mac Pro running OS X 10.6.8 Server - 8GB RAM - dedicated to running Retrospect 9 - Tandberg Data LTO-4 HH external drive, SCSI interface - bonded (2x) dedicated gigabit ethernet - latest updates across-the-board Backup Client Machine Details: - PowerMac running OS X 10.5.8 with Retrospect 6 client (via dedicated gigabit ethernet) - Mac Mini running OS X 10.6.8 with Retrospect 6 client (via dedicated gigabit ethernet) - Mac Mini running OS X 10.6.8 Server with Retrospect 9 client (via dedicated gigabit ethernet) - Mac Mini running OS X 10.7.4 with Retrospect 9 client (via dedicated gigabit ethernet) - Intel Mac Pro Running OS X Server 10.6.8 running Retrospect 9 Server (via internal SATA HD) Related (Non-Retrospect) Speed Details - network is gigabit ethernet (managed by a stack of gigabit switches) - all machines on network have dedicated gigabit ethernet runs back to switch stack - network is capable of sustaining real world gigabit speeds of 75+ MB/sec — for most machines on network, HD speed is the limiting factor - HDs in testing range from (real world) speeds of 60+ MB/sec (low-end) to 95+ MB/sec (high-end) - Tandberg Data LTO-4 drive (when using Tandberg's diagnostic tools) averages 73 MB/sec (which is just shy of it's "listed" speed) Based on the above, using the low-end configuration for my baseline — ie. the machine with the 60 MB/sec HD, since this is the "limiting factor" in my testing — backups to the LTO4 *should* be capable of 3.5+ GB/min (again, all these speeds are "real world" in terms of, that's what I know I can typically get on my network, mostly via AFP, which has a lot of overhead in itself). But this isn't taking into account any overhead resulting from Retrospect. So, here are the details surrounding Retrospect 9… Test Backup Folder: - 15GB backup of various files (image files, movie files, InDesign files, Illustrator files, etc. Retrospect 9 backup scripts: - basic backup of all files within the designated folder, no encryption, software compression, thorough verification - basic backup of all files within the designated folder, no encryption, hardware compression, thorough verification - basic backup of all files within the designated folder, no encryption, no compression, thorough verification Average Results: - about 30 minutes to back up the 15GB @ 950 MB/minute (software compression - 31% compression) - about 15 minutes to back up 15GB @ 2.5 GB/minute (hardware compression - Retrospect did not report hardware compression results) - about 15 minutes to back up 15GB @ 2.5 GB/minute (no compression) - time include 3-4 minutes, on average, of idle/loading/preparing time of the drive during each backup On the high-end of the testing (no compression or hardware compression), both the backup via the server's internal drive and the backup via the Mac Mini (running Retrospect 9) over gigabit ethernet averaged 2.9 GB/minute, and on the low-end — the Mac Mini with the 60 MB/sec HD — it averaged 2.1 GB/minute. These are Retrospect's reported speeds. So (and I'm not complaining, since I know Retrospect has a lot of overhead as well), I'd say, within my backup environment, my limiting factor would be Retrospect. That all said — the main thing I'm wondering is if it's normal for software compression to take so long compared to hardware compression or no compression? Any advise, opinions, etc. on the above would be appreciated. Thanks, Kristin. Quote Link to comment Share on other sites More sharing options...
Lennart_T Posted July 3, 2012 Report Share Posted July 3, 2012 Yes, software compression takes a lot of CPU. There was a discussion about the Windows version a while ago. The conclusion was to buy a special hardware compression card to speed up "software" compression. Otherwise it simply took too much time. Since disk space is so cheap today, there is no reason to use software compression for Disk Media Sets. For Tape Media Sets, the hardware compression is fast. Quote Link to comment Share on other sites More sharing options...
theboyk Posted July 3, 2012 Author Report Share Posted July 3, 2012 Yea, I'm dealing with tape backups — was hoping to use hardware compression, but because I can't encrypt the data AND use hardware compression, was really hoping that I could use software compression. But, considering the 2x timeframe for software compression, it looks like I'll be better off with no compression. What kinda of "card" would be used for speeding up software compression? Would it also allow for encryption of the data? Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.