On 12/18/20 4:26 PM, Philip Pemberton via Bacula-users wrote:
Hi all, I'm trying to improve the performance of my Bacula backups. I have a configuration with two machines: - "A" - a small server running a few web services. (Intel Celeron J1800 2.4GHz dual-core) - "B" - a 9TB NAS with a Quantum Superloader LTO-6 SAS tape robot (Intel Q6600 3GHz quad-core) My issues are twofold: - Backups of "B" are done by the local Bacula FD/SD/DIR and spooled onto disk to reduce shoe-shining. The spool limit is 50GB, on a solid-state disk. It takes about 6 minutes to fill the spool file, and between 5 and 7 to write it out to tape. This gives an effective data rate (quoted in the log) of about 50MB/s, but the tape write rate (again, from the log) is closer 100-120MB/s. - Backups from A to B take a long time to spool to disk, but the tape phase goes as fast as the local backup. Bacula reports about 7MB/sec. I assume something is slowing down the network traffic. I have a couple of questions -- - Re. local "B" backups. Bacula seems to be writing to the spool file, then dumping it to tape. There's no spooling happening when the tape is being written to. Is there any way I can set Bacula up to do "A/B" or "ping-pong" buffering, or something better than its current 50% duty cycle? Otherwise it seems my only
I don't think so. It is best to make the data spool as large as possible. Spooling can slow down jobs that are larger than the spool storage size.
- Re. slow transfers from "A" to "B". What can I do to speed up the network transfers between the two? I find that SMB and NFS from my workstation to/from "A" or "B" is quite fast, certainly far higher than the ~7MB/s I'm seeing (quoted in the Bacula log). I'm not expecting to hit 100MB/s, but I was expecting better than 7MB/s!
It is not likely a network issue. 'A' does not have a strong processor. When compression is enabled for a job, it is the client that performs the compression. Likewise for data encryption. Try disabling both compression and encryption for the 'A' job, if enabled.
Also, the rate is based on the total run time for the job, including de-spooling attributes at the end. If attribute de-spooling is taking a long time, then database performance may be the bottleneck.
Both A and B are on the same gigabit network switch. "A" (small server) has an Intel 82574L Ethernet controller. "B" (NAS) has a Marvell 88E8056 Ethernet controller. Thanks, Phil phil...@philpem.me.uk _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
_______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
