vmstat -s
What does "revolutions of the clock hand" mean in a vmstat -s? I read somewhere that if "revolutions of the clock hand" > "uptime" then you have major problems. We have a system that just recently had "revolutions of the clock hand" = 10 and uptime was 2 days.... needless to say the shortly there after the system panicked. I can't find any documentation on the vmstat -s information. Thanks in advance.. Val -----Original Message----- From: Paul watson [mailto:watsop@russellcorp.co.uk] Sent: Friday, October 08, 1999 4:55 AM To: david_grove@health.state.ak.us Cc: informix-list@iiug.org Subject: Re: Swap Size for Solaris/IDS If you've got more than 48MB of memory as far as Solaris, assuming 2.4 or better, is concerned you don't not need to configure any disk based swap. If you've got enough memory in the box under normal operations you will never ever swap. To swap you have to be done to less than 2% of real memory. Having said that you need swap space so that programs can start. Any malloc or similiar the kernel will check that there is enough swap space available to swap to if required. This is a pure reservation process nothing else is done with this swap space, nothing is written to it. If you use shmat or similiar then the swap reservation process is not done, as the file pages are automagically brought into the program memory structures. You also need disk swap space if you want to do crash dump analysis, but when was the last time you analysed a kernel coredump to find out why it crashed. If it happens regularly then OK configure enough hard disk for it, solve the problem and then deallocate the area. You can have your swap space > memory in this instance what you can't have is it all in one place. So if you decide you want 6GB of swap just create 3 2GB swap areas. I tend to think of swap disk along the line of tempdbs and configure a few across a number of disks. The first swap area is the one used to store the crash dump if it happens and you have savecore enabled. The bigger your swap the bigger the filesystem cache as Solaris effectively treats all memory and swap, except for the kernel areas which can't be paged anyway, as file system cache. But vnode pages would never paged out to disk so you only really need to consider anon pages. But this box is only running Informix. Adrian tends to recommend large areas of swap space for database servers but he has an Oracle background and admits is Informix is not strong. The reason is when the application reserves swap space it does so from the back of virtual memory ie disk based swap. As this area is used up it reaches a point where all the disk based swap has been reserved and now the kernel is forced to reserve memory from real RAM. This is a bad thing!! How big is still the question. I'd start at about 1.5G and monitor it's usage. The only accurate measure of swap usage under Solaris is from vmstat -s, all other swap options or commands do not report what you think they do. As aside Solaris vmstat/mpstat etc do not accurately measure machine load, AFAIK it is the monitoring of context switching. This a problem within the kernel so using adb to get the information doesn't help. The problem get worse the more CPUs you have. It is supposed to be fixed in Solaris 2.8 Paul Watson WF Software Ltd Tel: +44 1436 674729 Fax: +44 1436 678729 www.wfsoftware.com/informix # If you broke it, hide the evidence >>> "David Grove" <david_grove@health.state.ak.us> 10/06/99 11:26PM >>> Folks, This topic is a mixture of Unix and Informix sysadmin. I posted here because I'm interested specifically (and only) in an Informix context. The questions are: Should swap partition be > 0? If so, how does one determine size? I have started some research in the newsgroups, as well as SunWorld, and Cockroft material. I have a long way to go. I'm finding the topic confusing, with seemingly contradictory recommendations ranging from "swap should be 3 times memory" to "don't need swap if you have more than several 100 MB memory". Background: We are using IDS 7.3 (likely soon to be 64-bit 7.30.FC7A-1) on Solaris 2.6 (definitely soon to be Solaris 7) on Sun UltraEnterprise 3000 with 3 CPUs, 2.5 GB memory, and 15 drives (distributed 3/6/6) on 3 controllers. The system is used exclusively for DSS (data warehouse). There is only a single IDS instance. At the moment, the warehouse is embryonic, being <1 GB. We expect it to be at least a year before size doubles, thus for near term, it is feasible to max out buffers (512K @ 2K each = 1 GB mem for buffers) and make everything resident. This still leaves decent amount DSS memory, say 1 GB, for various pools (sorts, etc.). I would like to know how to size swap for the upcoming Solaris 7 install. It would seem right off the bat that, if we do have swap, we certainly can't even have 1X amount of memory, since 2.5GB is already greater than the max allowed Solaris swap size of 2GB. What are the implications of swap being < RAM? It would seem to me that swap space becomes less significant as the ratio of swap to memory decreases. Of course, that can't be completely correct, if the advocates of small swap spaces are valid. Bottom line question: What size should I make swap, and what factors determine it? Thank you for any thoughts or comments. Regards, David Grove ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. This footnote also confirms that this email message has been swept by MIMEsweeper for the presence of computer viruses. www.mimesweeper.com **********************************************************************