Re: Hardware-assisted compression: not CPU-efficient?
I think your test was too small. I did not see any meaningful differences among your results. I'd go for test data of at least 100x in size. -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Yifat Oren Sent: Wednesday, December 08, 2010 12:22 PM To: IBM-MAIN@bama.ua.edu Subject: Re: Hardware-assisted compression: not CPU-efficient? Pardon my bringing back an old thread, but - I wanted to see how much better is the COMPRESS option over the HWCOMPRESS in regards to CPU time and was pretty surprised when my results suggested that HWCOMPRESS is persistently more efficient (both CPU and channel utilization -wise) than COMPRESS: DFDSS DUMP with OPT(4) of a VSAM basic format to disk (basic format): STEPNAME PROCSTEPRC EXCP CONNTCBSRB CLOCK DUMP-HWCOMPRESS 00 14514 93575.25.072.3 output was 958 cyls. DUMP-COMPRESS00 14819 92326.53.072.5 output was 978 cyls. DUMP-NOCOMP 00 15283 103K.13.082.4 output was 1,017 cyls. DFDSS DUMP with OPT(4) of a PS basic format to disk (basic format): STEPNAME PROCSTEPRC EXCP CONNTCBSRB CLOCK DUMP-HWCOMPRESS 00 13317 154K.44.196.2 output was 877 cyls. DUMP-COMPRESS00 14692 157K.68.195.1 output was 969 cyls. DUMP-NOCOMP 00 35827 238K.14.217.9 output was 2,363 cyls. Running on a 2098-I04. DFSMSDSS V1R09.0. So, how come I get different results than the original poster? The test data was database-type data sets.. Best Regards, Yifat -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Andrew N Wilt Sent: Friday, December 03, 2010 1:45 AM To: IBM-MAIN@bama.ua.edu Subject: Re: Hardware-assisted compression: not CPU-efficient? Ron, Thank you for the good response. It is true that the DFSMSdss COMPRESS keyword and HWCOMPRESS keyword do not perform the same types of compression. Like Ron said, the COMPRESS keyword is using a Huffman encoding technique, and works amazing for repeated bytes (just the types of things you see on system volumes). The HWCOMPRESS keyword utilizes a dictionary based method, and works well, supposedly, on customer type data. The CPU utilization of the HWCOMPRESS (dictionary based) is indeed larger due to what it is doing. So you should choose the type of compression that suits your CPU utilization needs and data type. It was mentioned elsewhere in this thread about using the Tape Hardware compaction. If you have it available, that's what I would go for. The main intent of the HWCOMPRESS keyword was to provide the dictionary based compression for the cases where you were using the software encryption, and thus couldn't utilize the compaction of the tape device. Thanks, Andrew Wilt IBM DFSMSdss Architecture/Development Tucson, Arizona -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html NOTICE: This electronic mail message and any files transmitted with it are intended exclusively for the individual or entity to which it is addressed. The message, together with any attachment, may contain confidential and/or privileged information. Any unauthorized review, use, printing, saving, copying, disclosure or distribution is strictly prohibited. If you have received this message in error, please immediately advise the sender by reply email and delete all copies. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Pardon my bringing back an old thread, but - I wanted to see how much better is the COMPRESS option over the HWCOMPRESS in regards to CPU time and was pretty surprised when my results suggested that HWCOMPRESS is persistently more efficient (both CPU and channel utilization -wise) than COMPRESS: DFDSS DUMP with OPT(4) of a VSAM basic format to disk (basic format): STEPNAME PROCSTEPRC EXCP CONNTCBSRB CLOCK DUMP-HWCOMPRESS 00 14514 93575.25.072.3 output was 958 cyls. DUMP-COMPRESS00 14819 92326.53.072.5 output was 978 cyls. DUMP-NOCOMP 00 15283 103K.13.082.4 output was 1,017 cyls. DFDSS DUMP with OPT(4) of a PS basic format to disk (basic format): STEPNAME PROCSTEPRC EXCP CONNTCBSRB CLOCK DUMP-HWCOMPRESS 00 13317 154K.44.196.2 output was 877 cyls. DUMP-COMPRESS00 14692 157K.68.195.1 output was 969 cyls. DUMP-NOCOMP 00 35827 238K.14.217.9 output was 2,363 cyls. Running on a 2098-I04. DFSMSDSS V1R09.0. So, how come I get different results than the original poster? The test data was database-type data sets.. Best Regards, Yifat -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Andrew N Wilt Sent: Friday, December 03, 2010 1:45 AM To: IBM-MAIN@bama.ua.edu Subject: Re: Hardware-assisted compression: not CPU-efficient? Ron, Thank you for the good response. It is true that the DFSMSdss COMPRESS keyword and HWCOMPRESS keyword do not perform the same types of compression. Like Ron said, the COMPRESS keyword is using a Huffman encoding technique, and works amazing for repeated bytes (just the types of things you see on system volumes). The HWCOMPRESS keyword utilizes a dictionary based method, and works well, supposedly, on customer type data. The CPU utilization of the HWCOMPRESS (dictionary based) is indeed larger due to what it is doing. So you should choose the type of compression that suits your CPU utilization needs and data type. It was mentioned elsewhere in this thread about using the Tape Hardware compaction. If you have it available, that's what I would go for. The main intent of the HWCOMPRESS keyword was to provide the dictionary based compression for the cases where you were using the software encryption, and thus couldn't utilize the compaction of the tape device. Thanks, Andrew Wilt IBM DFSMSdss Architecture/Development Tucson, Arizona -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Tony, Then the misunderstanding is that Compression Services as called by DFSMSdfp and DFSMSdss with HWCOMPRESS uses an LZW compression scheme, while DFSMShsm and DFSMSdss with the COMPRESS keyword use Huffman technique. The Asymmetric cost of HWCOMPRESS I was referring to, and that apparently confused you, is the same differential for LZW that you mention below. I suspected that you did not know the difference in encoding techniques, which is why I pointed it out. I'm aware of customers with Disaster Recovery schemes that rely on restores from DFSMSdss back-ups, and spend the first 12-24 hours of the DR drill restoring data from port and channel constrained cartridge drives. HWCOMPRESS would relieve that situation and potentially speed up the restore process by 50% or more without creating a CPU bottleneck that may occur with COMPRESS. My own findings a few years ago were that dumping to disk in the absence of channel and hardware buffer saturation (i.e. disk output) using COMPRESS runs slower than NOCOMPRESS. A decade and a half ago ago when I looked at this with ASTEX trace and GTFPARS it looked like read/write overlap was disabled when COMPRESS was used. I have not run any tests with HWCOMPRESS. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Tony Harminc > Sent: Thursday, December 02, 2010 4:51 PM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > On 2 December 2010 18:20, Ron Hawkins wrote: > > Tony, > > > > You are surprised, and then you explain your surprise by agreeing with me. > > I'm confused. > > Well now I'm confused; I'm not sure how I did what you say. > > > I'm not sure if you realized that the Huffman encoding technique used by > > DFMSdss COMPRESS keyword is not a dictionary based method, and has a > > symmetrical CPU cost for compression and decompression. > > No, I didn't know anything about the compression methods triggered by > these two keywords until this thread. But I do know to some extent how > both Huffman and the LZW-style dictionary compression schemes work, > and that there is a differential between encoding and decoding speed > when an inherently adaptive scheme like LZW is used, vs a usually > static Huffman scheme. > > But I'm afraid I'm missing your point. You said that the saving in > hardware assisted compression is in decompression, and I took this to > be a claim that hardware assisted decompression is somehow speeded up > - when compared to a plain software implementation - relatively more > than is compression, and I said that I doubt that that is the case. > But if it is indeed the case under some circumstances, then I don't > see why most shops would care in most cases. > > > Finally, as I mentioned in another email, there may be intrinsic Business > > Continuance value in taking advantage of the asymmetric CPU cost to speed up > > local recovery of an application, or Disaster Recovery that is based on > > DFSMSdss restores. An improvement in Recovery time may be worth the > > increased cost of the backup. > > It's certainly possible, but I think it is unlikely to be the common case. > > Tony H. > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
On 2 December 2010 18:20, Ron Hawkins wrote: > Tony, > > You are surprised, and then you explain your surprise by agreeing with me. > I'm confused. Well now I'm confused; I'm not sure how I did what you say. > I'm not sure if you realized that the Huffman encoding technique used by > DFMSdss COMPRESS keyword is not a dictionary based method, and has a > symmetrical CPU cost for compression and decompression. No, I didn't know anything about the compression methods triggered by these two keywords until this thread. But I do know to some extent how both Huffman and the LZW-style dictionary compression schemes work, and that there is a differential between encoding and decoding speed when an inherently adaptive scheme like LZW is used, vs a usually static Huffman scheme. But I'm afraid I'm missing your point. You said that the saving in hardware assisted compression is in decompression, and I took this to be a claim that hardware assisted decompression is somehow speeded up - when compared to a plain software implementation - relatively more than is compression, and I said that I doubt that that is the case. But if it is indeed the case under some circumstances, then I don't see why most shops would care in most cases. > Finally, as I mentioned in another email, there may be intrinsic Business > Continuance value in taking advantage of the asymmetric CPU cost to speed up > local recovery of an application, or Disaster Recovery that is based on > DFSMSdss restores. An improvement in Recovery time may be worth the > increased cost of the backup. It's certainly possible, but I think it is unlikely to be the common case. Tony H. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Ron, Thank you for the good response. It is true that the DFSMSdss COMPRESS keyword and HWCOMPRESS keyword do not perform the same types of compression. Like Ron said, the COMPRESS keyword is using a Huffman encoding technique, and works amazing for repeated bytes (just the types of things you see on system volumes). The HWCOMPRESS keyword utilizes a dictionary based method, and works well, supposedly, on customer type data. The CPU utilization of the HWCOMPRESS (dictionary based) is indeed larger due to what it is doing. So you should choose the type of compression that suits your CPU utilization needs and data type. It was mentioned elsewhere in this thread about using the Tape Hardware compaction. If you have it available, that's what I would go for. The main intent of the HWCOMPRESS keyword was to provide the dictionary based compression for the cases where you were using the software encryption, and thus couldn't utilize the compaction of the tape device. Thanks, Andrew Wilt IBM DFSMSdss Architecture/Development Tucson, Arizona IBM Mainframe Discussion List wrote on 12/02/2010 04:20:15 PM: > From: > > Ron Hawkins > > To: > > IBM-MAIN@bama.ua.edu > > Date: > > 12/02/2010 04:21 PM > > Subject: > > Re: Hardware-assisted compression: not CPU-efficient? > > Tony, > > You are surprised, and then you explain your surprise by agreeing with me. > I'm confused. > > I'm not sure if you realized that the Huffman encoding technique used by > DFMSdss COMPRESS keyword is not a dictionary based method, and has a > symmetrical CPU cost for compression and decompression. > > Finally, as I mentioned in another email, there may be intrinsic Business > Continuance value in taking advantage of the asymmetric CPU cost to speed up > local recovery of an application, or Disaster Recovery that is based on > DFSMSdss restores. An improvement in Recovery time may be worth the > increased cost of the backup. > > Ron > > > -Original Message- > > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On > Behalf Of > > Tony Harminc > > Sent: Thursday, December 02, 2010 9:09 AM > > To: IBM-MAIN@bama.ua.edu > > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > > > On 2 December 2010 05:53, Ron Hawkins > wrote: > > > Johnny, > > > > > > The saving in hardware assisted compression is in decompression - when > you > > read it. Look at what should be a much lower CPU cost to decompress the > files > > during restore and decide if the speed of restoring the data concurrently > is > > worth the increase in CPU required to back it up in the first place. > > > > I am a little surprised at this. Certainly for most of the current > > dynamic dictionary based algorithms (and many more as well), > > decompression will always, except in pathological cases, be a good > > deal faster than compression. This is intuitively obvious, since the > > compression code must not only go through the mechanics of > > transforming input data into the output codestream, but must do it > > with some eye to actually compressing as best it can with the > > knowledge available to it, rather than making things worse. The > > decompression simply takes what it is given, and algorithmically > > transforms it back with no choice. > > > > Whether a hardware assisted - which in this case means one using the > > tree manipulation instructions - decompression is disproportionately > > faster than a similar compression, I don't know, but I'd be surprised > > if it's much different. > > > > But regardless, surely it is a strange claim that an installation > > would use hardware assisted compression in order to make their > > restores faster, particularly at the expense of their dumps. What > > would be the business case for such a thing? How many installations do > > restores on any kind of regular basis? How many have a need to have > > them run even faster than they do naturally when compared to the > > dumps? > > > > Tony H. > > > > -- > > For IBM-MAIN subscribe / signoff / archive access instructions, > > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > > Search the archives at http://bama.ua.edu/archives/ibm-main.html > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Ted, I think that why us DASD vendors invented Concurrent Copy, Snapshot, Shadowimage, Timefinder and FlashCopy. The backup is done relatively quickly, and copying the backup to tape can be completed outside the Business Critical Path. I'm not suggesting for a moment that everyone uses these products. I did say that the increase cost and time for backup needs to evaluated against any improvement in restoration time with hardware compression. Thank you to all those that reinforced the need for this evaluation in their response. > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Ted MacNEIL > Sent: Thursday, December 02, 2010 1:14 PM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > >opposed to back-up duration which is usually outside of > any business critical path. > > It shouldn't be, especially if back-ups have to complete before sub-systems > can come up. > > If we ran out of window, we had senior IT management and business contacts > decide which was more critical: back-up; availability. > > Sometimes, like during the Christmas shopping season the decision was > availability. > > But, that was mortgaging the future. > Recovering without back-ups during your peak season takes longer than during > 'normal' times. > > We never had to run recovery when we made the decision, but I was glad I > didn't have the responsibility to make the choice. > > Back-ups are insurance premiums. > If you pay and nothing happens, it's a business expense. > If you don't pay and something happens, it may be a career event! > > - > Ted MacNEIL > eamacn...@yahoo.ca > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Tony, You are surprised, and then you explain your surprise by agreeing with me. I'm confused. I'm not sure if you realized that the Huffman encoding technique used by DFMSdss COMPRESS keyword is not a dictionary based method, and has a symmetrical CPU cost for compression and decompression. Finally, as I mentioned in another email, there may be intrinsic Business Continuance value in taking advantage of the asymmetric CPU cost to speed up local recovery of an application, or Disaster Recovery that is based on DFSMSdss restores. An improvement in Recovery time may be worth the increased cost of the backup. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Tony Harminc > Sent: Thursday, December 02, 2010 9:09 AM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > On 2 December 2010 05:53, Ron Hawkins wrote: > > Johnny, > > > > The saving in hardware assisted compression is in decompression - when you > read it. Look at what should be a much lower CPU cost to decompress the files > during restore and decide if the speed of restoring the data concurrently is > worth the increase in CPU required to back it up in the first place. > > I am a little surprised at this. Certainly for most of the current > dynamic dictionary based algorithms (and many more as well), > decompression will always, except in pathological cases, be a good > deal faster than compression. This is intuitively obvious, since the > compression code must not only go through the mechanics of > transforming input data into the output codestream, but must do it > with some eye to actually compressing as best it can with the > knowledge available to it, rather than making things worse. The > decompression simply takes what it is given, and algorithmically > transforms it back with no choice. > > Whether a hardware assisted - which in this case means one using the > tree manipulation instructions - decompression is disproportionately > faster than a similar compression, I don't know, but I'd be surprised > if it's much different. > > But regardless, surely it is a strange claim that an installation > would use hardware assisted compression in order to make their > restores faster, particularly at the expense of their dumps. What > would be the business case for such a thing? How many installations do > restores on any kind of regular basis? How many have a need to have > them run even faster than they do naturally when compared to the > dumps? > > Tony H. > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
>opposed to back-up duration which is usually outside of any business critical path. It shouldn't be, especially if back-ups have to complete before sub-systems can come up. If we ran out of window, we had senior IT management and business contacts decide which was more critical: back-up; availability. Sometimes, like during the Christmas shopping season the decision was availability. But, that was mortgaging the future. Recovering without back-ups during your peak season takes longer than during 'normal' times. We never had to run recovery when we made the decision, but I was glad I didn't have the responsibility to make the choice. Back-ups are insurance premiums. If you pay and nothing happens, it's a business expense. If you don't pay and something happens, it may be a career event! - Ted MacNEIL eamacn...@yahoo.ca -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Hal Merritt Sent: Friday, 3 December 2010 6:44 AM To: IBM-MAIN@bama.ua.edu Subject: Re: Hardware-assisted compression: not CPU-efficient? Conversely, sometimes it is hard to get the backups all done in a low activity window, so one might compromise in favor of faster backups even at the expense of more CPU consumption. Depending on shop's strategy, getting a logically consistent PIT copy just might put the backups in the business critical path. That is, all have to complete before the next business day starts. --- Doesn't have to be if you combine the backups with hardware vendor replication technologies such as SHADOWIMAGE, TIMEFINDER and FLASHCOPY. Read how Innovation's FDRINSTANT solution gets around the issue of taking backups off the critical path: http://www.innovationdp.fdr.com/products/fdrinstant/ Stephen Mednick Computer Supervisory Services Sydney, Australia Asia/Pacific representatives for Innovation Data Processing, Inc. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Conversely, sometimes it is hard to get the backups all done in a low activity window, so one might compromise in favor of faster backups even at the expense of more CPU consumption. Depending on shop's strategy, getting a logically consistent PIT copy just might put the backups in the business critical path. That is, all have to complete before the next business day starts. -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Ron Hawkins Sent: Thursday, December 02, 2010 12:58 PM To: IBM-MAIN@bama.ua.edu Subject: Re: Hardware-assisted compression: not CPU-efficient? Gil, I was thinking that a faster restore would be have some value as a reduction in recovery time, as opposed to back-up duration which is usually outside of any business critical path. This would have value in business continuance whether it was a small application recovery or a full disaster recovery situation. I don't think the frequency of recovery is a factor in this case. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Paul Gilmartin > Sent: Thursday, December 02, 2010 6:09 AM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > On Thu, 2 Dec 2010 02:53:17 -0800, Ron Hawkins wrote: > > > >The saving in hardware assisted compression is in decompression - when you > read it. Look at what should be a much lower CPU cost to decompress the files > during restore and decide if the speed of restoring the data concurrently is > worth the increase in CPU required to back it up in the first place. > > > So if you restore more frequently than you backup, you come out ahead? > > -- gil > -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html NOTICE: This electronic mail message and any files transmitted with it are intended exclusively for the individual or entity to which it is addressed. The message, together with any attachment, may contain confidential and/or privileged information. Any unauthorized review, use, printing, saving, copying, disclosure or distribution is strictly prohibited. If you have received this message in error, please immediately advise the sender by reply email and delete all copies. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Gil, I was thinking that a faster restore would be have some value as a reduction in recovery time, as opposed to back-up duration which is usually outside of any business critical path. This would have value in business continuance whether it was a small application recovery or a full disaster recovery situation. I don't think the frequency of recovery is a factor in this case. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Paul Gilmartin > Sent: Thursday, December 02, 2010 6:09 AM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > On Thu, 2 Dec 2010 02:53:17 -0800, Ron Hawkins wrote: > > > >The saving in hardware assisted compression is in decompression - when you > read it. Look at what should be a much lower CPU cost to decompress the files > during restore and decide if the speed of restoring the data concurrently is > worth the increase in CPU required to back it up in the first place. > > > So if you restore more frequently than you backup, you come out ahead? > > -- gil > -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Martin, Except for when the compression assist instructions were in millicode on the G4 and G5, the hardware compression from Compression Services has always had am asymmetric cost for DFDMS compression. I remember some early documentation from IBM when it was first introduced in DFSMS that quoted 12 instructions per byte to compress, and two instructions per byte to decompress. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Martin Packer > Sent: Thursday, December 02, 2010 3:36 AM > To: IBM-MAIN@bama.ua.edu > Subject: Re: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > Ron, is it generally the case that CPU is saved on read? I'm seeing QSAM > with HDC jobsteps showing very high CPU. But then they seem to both write > and read. Enough CPU to potentially suffer from queuing. > > (And, yes, I know you were talking about a different category of HDC > usage.) > > Martin Packer, > Mainframe Performance Consultant, zChampion > Worldwide Banking Center of Excellence, IBM > > +44-7802-245-584 > > email: martin_pac...@uk.ibm.com > > Twitter / Facebook IDs: MartinPacker > > > > > > Unless stated otherwise above: > IBM United Kingdom Limited - Registered in England and Wales with number > 741598. > Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU > > > > > > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Unfortunately, IBM, et. al. *DO NOT* bill on elapsed time. More CPU used for Dump is less CPU available for productive work, or worse yet, a bigger software bill! Increased CPU time to do the dump does not necessarily mean that the elapsed time is longer. In fact, by compressing the data, I would expect that the time required to write it out (the I/O time) would be less. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
On Thu, 2 Dec 2010 12:09:23 -0500, Tony Harminc wrote: >On 2 December 2010 05:53, Ron Hawkins wrote: >> >> The saving in hardware assisted compression is in >>decompression - when you read it. Look at what should be a >>much lower CPU cost to decompress the files during restore >>and decide if the speed of restoring the data concurrently is >>worth the increase in CPU required to back it up in the first place. > >I am a little surprised at this > >But regardless, surely it is a strange claim that an installation >would use hardware assisted compression in order to make their >restores faster, particularly at the expense of their dumps. Increased CPU time to do the dump does not necessarily mean that the elapsed time is longer. In fact, by compressing the data, I would expect that the time required to write it out (the I/O time) would be less. -- Tom Marchant -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
On 2 December 2010 05:53, Ron Hawkins wrote: > Johnny, > > The saving in hardware assisted compression is in decompression - when you > read it. Look at what should be a much lower CPU cost to decompress the files > during restore and decide if the speed of restoring the data concurrently is > worth the increase in CPU required to back it up in the first place. I am a little surprised at this. Certainly for most of the current dynamic dictionary based algorithms (and many more as well), decompression will always, except in pathological cases, be a good deal faster than compression. This is intuitively obvious, since the compression code must not only go through the mechanics of transforming input data into the output codestream, but must do it with some eye to actually compressing as best it can with the knowledge available to it, rather than making things worse. The decompression simply takes what it is given, and algorithmically transforms it back with no choice. Whether a hardware assisted - which in this case means one using the tree manipulation instructions - decompression is disproportionately faster than a similar compression, I don't know, but I'd be surprised if it's much different. But regardless, surely it is a strange claim that an installation would use hardware assisted compression in order to make their restores faster, particularly at the expense of their dumps. What would be the business case for such a thing? How many installations do restores on any kind of regular basis? How many have a need to have them run even faster than they do naturally when compared to the dumps? Tony H. -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
On Thu, 2 Dec 2010 16:29:56 +0200, Yifat Oren wrote: > >I was under the impression that for DFDSS DUMP, COMPRESS and HWCOMPRESS are >synonymous; > >Are you saying they are not? > > Yes, they are not synonymous. HWCOMPRESS uses the CMPSC instruction (dictionary-based compression). COMPRESS uses RLE (run-length encoding). Norbert Friemel -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
"Yifat Oren" wrote in message news:<3d0c19e6913742b282eeb9a7c4ae3...@yifato>... > Hi Johnny, > > I was under the impression that for DFDSS DUMP, COMPRESS and HWCOMPRESS are > synonymous; > > Are you saying they are not? > > > If you are writing to tape why not use the drive compaction(DCB=TRTCH=COMP) > instead? Because he is trying to lower channel utilization, so he must compress before sending it over the channel. Kees. For information, services and offers, please visit our web site: http://www.klm.com. This e-mail and any attachment may contain confidential and privileged material intended for the addressee only. If you are not the addressee, you are notified that no part of the e-mail or any attachment may be disclosed, copied or distributed, and that any other action related to this e-mail or attachment is strictly prohibited, and may be unlawful. If you have received this e-mail by error, please notify the sender immediately by return e-mail, and delete this message. Koninklijke Luchtvaart Maatschappij NV (KLM), its subsidiaries and/or its employees shall not be liable for the incorrect or incomplete transmission of this e-mail or any attachments, nor responsible for any delay in receipt. Koninklijke Luchtvaart Maatschappij N.V. (also known as KLM Royal Dutch Airlines) is registered in Amstelveen, The Netherlands, with registered number 33014286 -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Hi Johnny, I was under the impression that for DFDSS DUMP, COMPRESS and HWCOMPRESS are synonymous; Are you saying they are not? If you are writing to tape why not use the drive compaction(DCB=TRTCH=COMP) instead? Best Regards, Yifat -Original Message- From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of Johnny Luo Sent: יום ה 02 דצמבר 2010 12:13 To: IBM-MAIN@bama.ua.edu Subject: Hardware-assisted compression: not CPU-efficient? Hi, DSS DUMP supports COMPRESS/HWCOMPRESS keyword and I found out in my test that HWCOMPRESS costs more CPU than COMPRESS. Is it normal? Currently we're dumping huge production data to tape and in order to alleviate the tape channel utilization we need to compress the data before writing to tape. It works well but the cpu usage is a problem cause we have many such backup jobs running simultaneously. If hardware-assisted compression cannot reduce the cpu overhead, I will consider using resource group to cap those jobs. Best Regards, Johnny Luo -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
On Thu, 2 Dec 2010 02:53:17 -0800, Ron Hawkins wrote: > >The saving in hardware assisted compression is in decompression - when you >read it. Look at what should be a much lower CPU cost to decompress the files >during restore and decide if the speed of restoring the data concurrently is >worth the increase in CPU required to back it up in the first place. > So if you restore more frequently than you backup, you come out ahead? -- gil -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Hi Yes, it is a c library On 12/2/2010 1:26 PM, Johnny Luo wrote: Miklos, What do you mean by 'zlib'? Is it free on z/OS? Best Regards, Johnny Luo On Thu, Dec 2, 2010 at 8:10 PM, Miklos Szigetvari< miklos.szigetv...@isis-papyrus.com> wrote: Hi A few years ago I have tried with hardware compression, as we are using intensively the "zlib" library (http://www.ietf.org/rfc/rfc1950.txt) to compress/expand . Never got a proper answer, and till now not clear, in which case would bring the hardware compression some CPU reduction On 12/2/2010 12:36 PM, Martin Packer wrote: Ron, is it generally the case that CPU is saved on read? I'm seeing QSAM with HDC jobsteps showing very high CPU. But then they seem to both write and read. Enough CPU to potentially suffer from queuing. (And, yes, I know you were talking about a different category of HDC usage.) Martin Packer, Mainframe Performance Consultant, zChampion Worldwide Banking Center of Excellence, IBM +44-7802-245-584 email: martin_pac...@uk.ibm.com Twitter / Facebook IDs: MartinPacker Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Miklos, What do you mean by 'zlib'? Is it free on z/OS? Best Regards, Johnny Luo On Thu, Dec 2, 2010 at 8:10 PM, Miklos Szigetvari < miklos.szigetv...@isis-papyrus.com> wrote: >Hi > > A few years ago I have tried with hardware compression, as we are using > intensively the "zlib" library (http://www.ietf.org/rfc/rfc1950.txt) to > compress/expand . > Never got a proper answer, and till now not clear, in which case would > bring the hardware compression some CPU reduction > > > On 12/2/2010 12:36 PM, Martin Packer wrote: > >> Ron, is it generally the case that CPU is saved on read? I'm seeing QSAM >> with HDC jobsteps showing very high CPU. But then they seem to both write >> and read. Enough CPU to potentially suffer from queuing. >> >> (And, yes, I know you were talking about a different category of HDC >> usage.) >> >> Martin Packer, >> Mainframe Performance Consultant, zChampion >> Worldwide Banking Center of Excellence, IBM >> >> +44-7802-245-584 >> >> email: martin_pac...@uk.ibm.com >> >> Twitter / Facebook IDs: MartinPacker >> >> >> >> >> >> Unless stated otherwise above: >> IBM United Kingdom Limited - Registered in England and Wales with number >> 741598. >> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU >> >> >> >> >> >> >> -- >> For IBM-MAIN subscribe / signoff / archive access instructions, >> send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO >> Search the archives at http://bama.ua.edu/archives/ibm-main.html >> >> >> > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html > -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Hi A few years ago I have tried with hardware compression, as we are using intensively the "zlib" library (http://www.ietf.org/rfc/rfc1950.txt) to compress/expand . Never got a proper answer, and till now not clear, in which case would bring the hardware compression some CPU reduction On 12/2/2010 12:36 PM, Martin Packer wrote: Ron, is it generally the case that CPU is saved on read? I'm seeing QSAM with HDC jobsteps showing very high CPU. But then they seem to both write and read. Enough CPU to potentially suffer from queuing. (And, yes, I know you were talking about a different category of HDC usage.) Martin Packer, Mainframe Performance Consultant, zChampion Worldwide Banking Center of Excellence, IBM +44-7802-245-584 email: martin_pac...@uk.ibm.com Twitter / Facebook IDs: MartinPacker Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Ron, is it generally the case that CPU is saved on read? I'm seeing QSAM with HDC jobsteps showing very high CPU. But then they seem to both write and read. Enough CPU to potentially suffer from queuing. (And, yes, I know you were talking about a different category of HDC usage.) Martin Packer, Mainframe Performance Consultant, zChampion Worldwide Banking Center of Excellence, IBM +44-7802-245-584 email: martin_pac...@uk.ibm.com Twitter / Facebook IDs: MartinPacker Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Re: Hardware-assisted compression: not CPU-efficient?
Johnny, The saving in hardware assisted compression is in decompression - when you read it. Look at what should be a much lower CPU cost to decompress the files during restore and decide if the speed of restoring the data concurrently is worth the increase in CPU required to back it up in the first place. Ron > -Original Message- > From: IBM Mainframe Discussion List [mailto:ibm-m...@bama.ua.edu] On Behalf Of > Johnny Luo > Sent: Thursday, December 02, 2010 2:13 AM > To: IBM-MAIN@bama.ua.edu > Subject: [IBM-MAIN] Hardware-assisted compression: not CPU-efficient? > > Hi, > > DSS DUMP supports COMPRESS/HWCOMPRESS keyword and I found out in my test > that HWCOMPRESS costs more CPU than COMPRESS. > > Is it normal? > > Currently we're dumping huge production data to tape and in order to > alleviate the tape channel utilization we need to compress the data before > writing to tape. It works well but the cpu usage is a problem cause we have > many such backup jobs running simultaneously. > > If hardware-assisted compression cannot reduce the cpu overhead, I will > consider using resource group to cap those jobs. > > Best Regards, > Johnny Luo > > -- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO > Search the archives at http://bama.ua.edu/archives/ibm-main.html -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
Hardware-assisted compression: not CPU-efficient?
Hi, DSS DUMP supports COMPRESS/HWCOMPRESS keyword and I found out in my test that HWCOMPRESS costs more CPU than COMPRESS. Is it normal? Currently we're dumping huge production data to tape and in order to alleviate the tape channel utilization we need to compress the data before writing to tape. It works well but the cpu usage is a problem cause we have many such backup jobs running simultaneously. If hardware-assisted compression cannot reduce the cpu overhead, I will consider using resource group to cap those jobs. Best Regards, Johnny Luo -- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html