On Fri, Sep 15, 2006 at 06:25:39AM +0200, Vidar Tyldum Hansen wrote: > Alain Fauconnet: > > > This page definitely is an excellent reference. If I may share my > > experience here: we're progressively moving away from hardware RAID. > > Most of our critical servers were IBM boxes, using IBM's ServeRAID h/w > > RAID controller (ips). We've had more downtime with these than with > > the boxes using kernel RAID: firmware bugs taking ages to be fixed, > > drives suddenly seen as 'defunct' by the controller for no apparent > > reason (they intensively tested OK on a non-RAID controller), very, > > *very* little information about what's going on reported by the > > controller and of course we were locked into using IBM disks only (the > > controller won't take anything else). > > Same experience here. I have 2 IBM servers and when the controller > decides a drive is defunct it will refuse to ever use it again. This > happens randomly. I put that disk into the other server and it works > flawlessly.
Hmmm... glad to see I'm not alone. How come these IBM RAID controllers seem relatively popular with all this? As for the so-called 'defunct' drives, I admittedly never had the curiosity to connect them to another ServeRAID. Interesting. What I have mistakenly done once is to use an Adaptec controller to initiate a low-level format of the drive. After that it was passing extensive R/W tests successfully on the machine that had the Adaptec controller, but the IBM box would not even detect that drive anymore (not defunct, just not there at all!). My guess is that the &[EMAIL PROTECTED] IBM controllers not only lock on firmware IDs, but the drives also come with some signature written in the data area that I had erased. Yuck. > > So my job at the moment involves moving disks between the servers :) > > > Long before that, we had a bunch of Acer servers using the Mylex > > DAC960. More or less the same story (except having to use only a > > particular brand of disks). > > I have had several controllers die on me, which is very bad for the RAID > state. Of course, the lack of cooling was probably contributing, but > that only goes to show that it's another level of complexity that can go > wrong. Agreed. These DAC960's have hung on me a few times too. Also I had to apply I can't recall how many firmware updates, which required downtime and don't come in through swup! > > > Of course this is just an example. Other h/w RAID controllers might be > > more reliable, more predictable, but so far I really trust kernel RAID > > more. > > Also, no need to much about with non-tsl-packaged drivers, and special > software. mdadm provides not only RAID creation and mangement, but also > alerting when a drive fails. Yup. > > I've made a policy not to use hardware RAID on Linux servers. > > > I'd use h/w RAID only on machines with extremely high I/O bandwidth > > requirements. > > I would like to show you a Norwegian benchmark of hardware and kernel > software RAID: > http://www.hwb.no/artikkel/15307 > > The words might not mean much to you, but take a look at the graphs. The > SIL-controller is just a plain SATA-controller: no hardware RAID. The > other controllers have hardware RAID, but are also tested using SW RAID. > > Reality may differ in special cases, but you shouldn't worry about > performance. That looks interesting although I miss the text. Since we're an international university, I'll try searching for a Norwegian guy around :-) Thanks. Greets, _A_ _______________________________________________ tsl-discuss mailing list [email protected] http://lists.trustix.org/mailman/listinfo/tsl-discuss
