Keith Brautigam wrote (2006/02/15): > Quinton Jansen wrote: > >What speeds are others getting when using an LTO-3 drive?
I take up interest just in direct writing speed to the tape reported by iostat, which I have typically about 60-80 MB/s, but still I think that it should be better, because 80 MB/s is native speed of LTO-3 drives without compression, so with compression it should be possible to go above 100 MB/s (look for example at http://h18006.www1.hp.com/products/quickspecs/11739_div/11739_div.html). > >If I'm reading the numbers correctly, 2GB in 13 minutes (10GB/hour) is way > >too slow (should be around 200G/hour). >From bacula report? There is a problem, that there are counted all times, like data spooling and attributes despooling (I suppose, that you do use spooling, because it is almost a "must" for LTO-3 drive). > >Feb 15 00:39:23 localhost kernel: scsi0 : Adaptec AIC79XX PCI-X SCSI HBA > >DRIVER, Rev 1.3.11 > >Feb 15 00:39:23 localhost kernel: <Adaptec 29320ALP Ultra320 SCSI > >adapter> > >Feb 15 00:39:23 localhost kernel: aic7901: Ultra320 Wide Channel > >A, SCSI Id=7, PCI-X 67-100Mhz, 512 SCBs It would be interesting to see the real agreed communication speed, beause I still have problem to run LTO-3 drive from HP with Adaptec using 320 MB/s (it is known bug). It works just with 160 MB/s, so I had to buy LSILogic, which does work with my LTO-3 over 320 MB/s very well. However, I did not get any speed differences :o( > My dedicated bacula server is a dual Athlon MP 2GzH system with 4 SATA > drives (add-in non-RAID 4 port Promise controller) and a Gig NIC running > Linux Ubuntu 5.10. I have the same version of bacula, and it is also > compiled from source. FreeBSD 6.0-STABLE, dual Intel Xeon 3.6 GHz, spooling is on 3ware 9500 with 7 drive RAID-0. > The main performance enhancement for me was switching to a system with > more CPU power. The second most important performance enhancement was > how I configured the systems four drives, so as to not be disk I/O bound This is what I already said before: Bacula has and will have more and more performance problems with high speed backup cofigurations. MD5 checksums should be either precomputed before writing data to the tape, or maybe there would help two writing threads instead of just one: One thread would read data and compute md5, the second thread would just try to write data to the tape. > (although that's still the slows part of the system). Two of the drives > are configured with software RAID1 for the OS etc. The other two (250GB > each) are configured as software RAID0 with XFS, and serve as a > dedicated data spool for bacula. I also spool attributes so that data > entry into the catalog does not slow writing to the tape. The second problem is that there can be just one spooling area, where one read + write RAID array may have performance problems even in very good configuration. There would be good to have some possibility to say, that spooling area is exclusively reserved for despooling process (or atleast more independent spooling areas, some reservation feature, ...). > not sure). I believe the advertised speed for LTO3 is around 60 MB/sec, 80 MB/s > but I think seeing a 44 MB/sec average in real world performance (on a > relatively old and cheap server) is pretty decent. However for LTO-3 it is not very satisfactory, because minimal physical speed accomodation for LTO-3 drives is 27 MB/s (HP) or 40 MB/s (IBM), so 44 MB/s is the lowver bound for IBM drives for best effectivity. It would be interesting to see, how many repositions is done during write. There is one value in HP drives (Repositions Per 100 MB), which can help in performance diagnostics. And I think that there are even better counters in IBM drives. > Also, I'm using an older SCSI controller card (160 not 320) and have > never tried to tune the drive with mt or any other utility. My experiences are not so good among 160 MB/s and 320 MB/s - I have the same performance in both cases for now. -- Rudolf Cejka <cejkar at fit.vutbr.cz> http://www.fit.vutbr.cz/~cejkar Brno University of Technology, Faculty of Information Technology Bozetechova 2, 612 66 Brno, Czech Republic ------------------------------------------------------- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642 _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
